Computed Tomography Signal and Images

When a CT scan is performed, ionizing radiation passes through the patient’s body, and a receptor records the strength of the x-ray emerging from the body. The attenuation of the x-ray beam varies, depending on the body tissues through which the beam passes. The degree of attenuation is essentially proportional to the physical density of the tissue (see Figure 15-1). Physically dense tissues such as bone attenuate x-ray to a greater degree than do less dense tissues. The relative density of tissues on CT can be remembered by considering the analogy of a glass of water into which different materials are placed. Air is least dense and remains above the water. Oil (fat) poured into the glass floats on the surface of the water because fat is less dense than water. Soft tissues such as a piece of meat sink to the bottom of the glass because they are slightly denser than water (remember that soft tissues consist of cells, which are composed predominantly of water). Bone is densest and falls to the bottom of the glass. On CT, a standardized density scale is applied, called the Hounsfield scale, measured in Hounsfield units. On this scale, air has a density of −1000 Hounsfield units, fat has a density of around −50 to −100 Hounsfield units, water has a density of 0 Hounsfield units, soft tissues have a density of around +50 to +100 Hounsfield units, and bone has a density of several hundred to +1000 Hounsfield units. A gray scale is assigned by density. The densest tissues are whitest, and the least dense are blackest, with intermediate densities assigned corresponding intermediate gray shades. Although the gray shade on CT can be adjusted to accentuate the appearance of certain tissues, a process called windowing, the relative brightness or darkness of tissues remains fixed because it depends on their fixed relative densities. Air is always blacker than bone. Pathologic processes can often be recognized on CT scan by changes in the density of tissues. For example, soft-tissue inflammation or edema results in an increase in the water content of adjacent fat, increasing its density. This results in an increase in the brightness of fat, called stranding. Abnormal fluid collections within soft tissues are usually visible on CT because they are lower in density than the normal soft tissue and appear darker. Air within soft-tissue infectious collections is black and is readily seen. Fractures in bone may be seen as discontinuities in the bright white cortex of bone.

Magnetic Resonance Signal and Appearance of Tissues in Resulting Images

A full description of the physics of MRI is beyond the scope of this text, as we focus on features important to emergency diagnosis. When MRI is performed, a magnetic field and radio frequency signal are applied, resulting in an emitted signal from tissue protons flipping their magnetic moment. The brightness of a tissue is therefore not proportional to physical density but rather proportional to proton richness and the characteristics of the applied magnetic field and radio signal, which can be varied to achieve different diagnostic effects (see Figure 15-1). Proton-rich tissues such as water can generate a strong signal; proton-poor tissues cannot generate a strong signal. Tissues devoid of protons, such as cortical bone and air, generate no signal.

A strong MR signal is assigned a bright (white) color on all MRI sequences. A weak signal is assigned black, with intermediate signal strengths receiving intermediate gray shades. The brightness of tissues on MR images therefore depends on the strength of the emitted signal from protons in response to a given pulse sequence (magnetic field and radio frequency manipulation). For example, air and calcified bone have a paucity of protons, emit no signal regardless of the magnetic field manipulation, and appear black on all MRI pulse sequences. Note this difference from CT, in which air and bone occupy opposite ends of the gray scale because of their widely different physical densities and x-ray attenuations.

Excluding air and cortical bone, other tissues have variable appearances on MRI depending on the character of the applied magnetic field and radio frequency signal, which can be manipulated in many ways. A given manipulation is called an MRI pulse sequence. Protons in tissues of various types respond differently to variations in the applied magnetic field and radio signal, depending on the other properties of the tissue. In response to some pulse sequences, water-rich tissues emit a strong signal and appear white. In response to other pulse sequences, they emit a weaker signal and appear black. Hematopoietic bone marrow is hyperintense to (brighter than) skeletal muscle on all pulse sequences. Fat in marrow may be bright (T1) or dark: T2, fast T2 with fat saturation, or short T1 inversion recovery (STIR)—see the later descriptions of these pulse sequences (Figure 15-2). Table 15-2 lists the appearance of various tissue types on MRI sequences.

Table 15-2 Tissue Appearances on Magnetic Resonance Imaging Sequences

Tissue	Appearance	Best Sequence
Air	Black on all sequences	Not applicable
Bone	Black on all sequences	• STIR (inversion recovery) • Fast T2 with fat saturation • T1
Articular cartilage	Variable	• STIR • Fast T2 with fat saturation • Gradient recall echo with fat saturation
Meniscus	Dark on all sequences	• Gradient recall echo T2 (T2*) • T1 • Spin–echo proton density
Labrum	Dark on all sequences	• T1 with intra-articular gadolinium • Gradient recall echo T2 (T2*)
Tendons or ligaments	Usually dark on all sequences except the anterior cruciate ligament, which has a striated appearance	• Gradient recall echo T2 (T2*) • STIR • Fast T2 with or without fat saturation
Muscle	Intermediate intensity on all sequences	• STIR • T1
Fat	Variable	• T1
Synovium	Invisible unless pathologically thickened	• T1 fat-saturated with IV gadolinium

We discuss the clinical applications of common pulse sequences later in this chapter.

Signal Strength and Image Resolution in Magnetic Resonance Imaging

An ideal MR image would have good signal-to-noise ratio and high spatial resolution. Unfortunately, these characteristics are inversely related because of the way that MR images are created (Figure 15-3). MR images are generated from the signals emitted from three-dimensional tissue volumes. The total signal from a given tissue volume is displayed as a single pixel, also called a volume element or “voxel” because of its three-dimensional source. The number of protons in a voxel is proportional to the volume of the voxel. The resolution of the image is inversely proportional to the voxel size. Using many small tissue volumes results in a high-resolution image composed of many small voxels. However, because the number of protons in each of these small volumes is also small, the signal from each voxel is weak compared with the background noise. The signal-to-noise ratio can be improved by increasing the size of the voxels, but this necessarily reduces the resolution of the image. The signal-to-noise ratio can be improved by moving the MRI coil (the source of the magnetic field) closer to the imaged body part, allowing high-resolution extremity images to be acquired while maintaining good signal strength by using dedicated extremity coils.^6-7

Technical Features of Computed Tomography and Magnetic Resonance Affecting Ability to Diagnose Fracture

CT and MR diagnoses of fracture rely on different imaging findings.

Modern CT scanners demonstrate the cortex of bone at very high spatial resolution. The cortex of bone is readily visible on CT because it consists of dense calcium, which attenuates the x-ray beam to a great extent, as described earlier. A fracture plane is a discontinuity in the cortex, usually filled with low-density hemorrhage, which attenuates the x-ray beam to a significantly lesser extent. A brief comparison with earlier CT scan technology highlights the mechanism for improved diagnostic accuracy of CT over time. Early CT scanners performed a series of single axial slices, with gaps between adjacent slices. A narrow fracture in the axial plane might fall between two adjacent slices and be missed. In addition, the thickness of the axial slices was as great as 5 mm with early scanners. Consequently, each displayed planar axial image represented the pooled data from several millimeters of anatomy in the z-axis. The low density of a fractured cortex might be obscured when it was “volume averaged” with data from the intact cortex in the adjacent several millimeters of anatomy. In comparison, modern multislice CT scanners perform helical acquisition of slices as thin as 0.625 mm, with no gaps between image slices (Figure 15-5). These form a three-dimensional volume of image data that can be displayed in axial, coronal, or sagittal planes. The probability of missing even a narrow fracture plane is thus greatly reduced.⁸

MRI demonstrates fractures through a different mechanism. The calcified bony cortex and trabecular bone are devoid of resonating protons, which are found predominantly in tissues containing fat or water. Consequently, the cortex and trabeculae produce essentially no signal and appear as black signal voids on all MRI pulse sequences. However, bone marrow contains fat, which produces high signal on some pulse sequences. In addition, in the presence of fracture, hemorrhage and edema in the marrow space provide resonating protons that produce a strong signal; this pathologic signal can be differentiated from normal marrow fat. Thus MRI provides an indirect diagnostic finding of fracture, rather than directly visualizing cortical fracture.

Nephrogenic Systemic Fibrosis and Gadolinium Contrast

Although the risk for contrast nephropathy from iodinated contrast used in CT is familiar to emergency physicians, gadolinium-associated nephrogenic systemic fibrosis may be less recognized. Nephrogenic systemic fibrosis (NSF) is a potentially fatal condition first linked to gadolinium contrast exposure in 2006.⁹ The condition results in tissue fibrosis clinically similar to scleroderma—sometimes localized, but in other cases involving multiple internal organs and progressing to death.

Contrast nephropathy from iodinated contrast agents leads to worsened renal insufficiency in patients with baseline renal impairment but has no important consequences in patients already receiving hemodialysis for chronic renal failure.¹⁰ In stark comparison, gadolinium-associated nephrogenic systemic fibrosis appears to pose the greatest risk to patients with acute and chronic renal insufficiency, including patients receiving chronic hemodialysis.¹¹ Because gadolinium contrast agents are largely renally excreted, with a half-life of approximately 90 minutes in patients with normal renal function, patients with low glomerular filtration rates (GFRs) are exposed to circulating gadolinium for longer periods.¹² For one common agent, gadodiamide, the half-life in patients with end-stage renal disease but not yet receiving dialysis is more than 34 hours and in peritoneal dialysis patients is more than 52 hours. Gadolinium agents are removed by hemodialysis, with approximately 65% to 78% cleared during a first hemodialysis session, 96% following a second session, and 99% following a third session. Peritoneal dialysis is poor at removing these agents; after 22 days of continuous peritoneal dialysis, only 69% of gadolinium contrast is excreted.¹²

Gadolinium-associated NSF is a grave but rare risk. Prince et al. retrospectively reviewed 74,124 patients receiving gadolinium-enhanced MRI over 10 years at two large medical centers and found no cases of biopsy-proven NSF after a standard gadolinium dose, 0.1 mmol/kg. Among patients receiving a high dose of gadolinium (between 0.2 and 0.4 mmol/kg), 15 of 8997 (0.17%) developed NSF, and all had an estimated GFR of less than 30 mL per minute. Of these patients, 11 had acute or acute-on-chronic renal insufficiency. Patients receiving chronic hemodialysis for end-stage renal disease and receiving high-dose gadolinium developed NSF at a rate of 0.4%. The highest incidence of NSF was observed in patients receiving high-dose gadolinium under two circumstances: 8.8% (10 of 114) in patients with a very low GFR (<15 mL/min) but not yet receiving hemodialysis and 19% (11 of 58) in those with acute renal failure and a creatinine increase of 0.5 mg/dL or greater in a 24-hour period. Hyperphosphatemia was also associated with increased NSF risk. No patient developed NSF if hemodialysis was performed within 24 hours of gadolinium exposure, and hemodialysis within 48 hours was associated with decreased risk. The risk of NSF appeared greatest in patients receiving Gadodiamide (GE Healthcare), while other gadolinium-based contrast agents were rarely associated with NSF.¹¹

For the emergency physician, several key points bear repeating. The risk for NSF is extremely low in patients receiving standard doses of gadolinium, regardless of renal function. The risk remains very small in those receiving high-dose gadolinium, except in patients with acute renal failure with an estimated GFR of less than 30 mL per minute. The risk is particularly high in patients with acute and worsening renal failure, and gadolinium should be avoided in this group whenever possible. Creatinine clearance, not measured serum creatinine, should be used to categorize renal function, and is calculated by a simple formula (Box 15-1) incorporating serum creatinine, patient age, gender, and body mass. Creatinine clearance calculators are readily available online, and some clinical laboratories now report creatinine clearance in addition to measured creatinine. An example illustrates the importance of using calculated creatinine clearance rather than measured serum creatinine. A measured creatinine of 1.0 mg/dL is within normal limits for most clinical laboratories. In a 70-kg, 35-year-old male, this corresponds to a (normal) calculated creatinine clearance of 90 mL per minute. However, in a 50-kg, 85-year-old female, this corresponds to a calculated creatinine clearance of 32 mL per minute, representing significant renal dysfunction.

When standard-dose gadolinium is administered, the risk for gadolinium-induced NSF is lower than the risk for death from iodinated contrast agents used in CT. Nonetheless, the American College of Radiology (ACR) now recommends informed consent before administration of gadolinium contrast for patients with moderate to end-stage renal disease.¹³ Patients receiving chronic hemodialysis should receive the lowest feasible gadolinium dose and should undergo dialysis as soon after contrast administration as possible. The ACR recommends hemodialysis within 2 hours after administration of gadolinium contrast for patients with renal failure,¹³ though this practice is not well supported by research evidence. As described earlier, peritoneal dialysis is ineffective at eliminating gadolinium, and these patients may be at particularly high risk from gadolinium; they should avoid exposure or undergo rapid hemodialysis following gadolinium adminstration.¹³ These are important considerations because dialysis patients may be at risk for conditions such as spinal epidural abscess, best diagnosed by MRI, resulting from hematogenous infection from dialysis catheters.¹⁴

Gadolinium Safety in Pregnancy

The effects of gadolinium on the developing human fetus are unknown. Gadolinium crosses the placenta from the maternal circulation to the fetal circulation, is filtered by the fetal kidney, and is then excreted into the amniotic fluid, where it may remain for a prolonged period. The ACR recommends against the routine use of gadolinium-based contrast agents in pregnancy. Instead, the ACR recommends that the decision to use gadolinium should be based on overwhelming potential benefit to the patient or fetus, and a discussion of risks and benefits with patient consent should be documented in the medical record.¹³

Allergies to Gadolinium Contrast Agents

Allergic reactions to gadolinium-based contrast are relatively rare (around 3%) but can occur, with an increased risk in patients with prior gadolinium reactions. Prior reactions to iodinated contrast are associated with increased risk for gadolinium reaction, around 6.3%. Asthma and allergies to other allergens are risk factors for gadolinium contrast reaction, with a two- to fourfold increase in risk.¹³

Magnetic Resonance Imaging in Patients With Pacemakers and Automatic Implantable Cardioverter–Defibrillators

Until recently MRI had long been considered contraindicated in patients with ferromagnetic foreign bodies, including pacemakers, automatic implantable cardioverter–defibrillators (AICDs), and other electronic devices. The magnetic field from MRI can cause programming changes to the devices, inhibition of pacing, rapid pacing, induction of ventricular fibrillation, heating of the device or leads, battery depletion, and device damage requiring replacement. However, accidental exposures without adverse events, and more than 230 published prospective cases of patients with pacemakers safely having undergone low-field MRI, suggest that with proper planning some patients with these devices may be able to undergo MRI if clinically necessary.¹⁵ In vitro studies of pacemakers and AICDs subjected to 1.5- to 3.0-tesla MRI found that no significant temperature change or permanent device damage resulted.^14-17 The ACR designates pacemakers and defibrillators as relative contraindications to MRI and states that MRI of patients with these devices should not be routine but may be considered case by case.¹³ Despite this, emergency physicians may find obtaining MRI in patients with these devices difficult because of fear of harm shared by the radiologist and the patient.

Magnetic Resonance Imaging in Patients With Intracranial Aneurysm Clips

Great care should be taken in subjecting patients with intracranial aneurysm clips to MRI, even musculoskeletal MRI of remote body regions. Titanium clips are safe for MRI. Other intracranial aneurysm clips can deflect when subjected to a magnetic field, resulting in serious injury or death. Nontitanium clips manufactured in 1995 or later are considered safe when the product labeling indicates MRI compatibility. Older clips may require proof of testing for ferromagnetic properties. Never assume that a patient with an aneurysm clip who has undergone prior MRI safely can do so again—variations in magnetic field characteristics and clip orientation in the field may create different effects on the clip. Significant injuries have been reported, including blindness.¹³

Other Magnetic Resonance Imaging Safety Considerations

MRI has several other potential injurious effects. First, MRI examinations result in high acoustic volume—more than 100 dB in some clinically used machines and up to 130 dB for experimental 4.7-tesla systems^18-19—loud enough to damage human hearing. The ACR recommends hearing protection be used by patients undergoing MRI.¹³

Second, MRI can induce current in conducting circuits, even when these loops are not connected to a device or power supply. Induced currents can lead to rapid heating, sometimes resulting in severe tissue injury. Thus wires such as external electrocardiogram monitor leads and even the foil backing material used in some cutaneous drug-delivery systems should be removed from the patient before MRI—it is not sufficient to disconnect wiring from a power source. If wiring cannot be removed, insulation or a heat sink such as an ice pack should be placed between the wire and the patient’s skin. Permanent neurologic injury from heating of implanted neurostimulator leads has been reported to the U.S. Food and Drug Administration. Patients themselves can form electrical circuits and should be instructed to avoid crossing arms and legs during MRI to lessen this risk. Nonferromagnetic skin staples can become heated in the magnetic field and can cause cutaneous burns; this can be avoided by applying an ice pack to the staple line to act as a heat sink. Magnetic heating of tattoos can lead to burning sensations and first-degree burns, again preventable by application of an ice pack before MRI.¹³ These injuries are rare and generally not serious but might interfere with completion of the MRI examination.^17,20-22 Patients should be warned of thermal effects of MRI and cautioned to report any heat sensation immediately. The MRI field strength influences thermal heating. MRI-compatible medical devices should not be imaged using MRI of a field strength different from that approved by the manufacturer because of risk for field-induced heating. In vitro studies of intrauterine devices including copper intrauterine devices suggest them to be safe, with no deflection, torque, heating, or image artifact.²³

Common Biases in Studies of Diagnostic Tests

Table 15-4 reviews common forms of limitation or bias occurring in studies evaluating diagnostic tests.²⁴

Table 15-4 Biases and Limitations in Studies of Diagnostic Tests

Limitation or Bias	Description or Example	Result
Lack of gold standard (see verification or workup bias)	No gold standard is available for the condition under consideration.	Without a valid diagnostic standard, true and false positive and negative results cannot be determined. Results and conclusions of the study cannot be rigorously evaluated for validity.
Incorporation bias	The test under evaluation is used as the gold standard, or the final diagnosis partly relies on this test result.	Results of the study are biased in favor of the apparent accuracy of the diagnostic test.
Lack of blinding	Results of other testing or patient characteristics are available to the researchers, potentially affecting their interpretation of the diagnostic test under consideration.	Results are biased by the physicians’ or researchers’ preconceptions about the accuracy of the diagnostic test.
Verification or workup bias	Not all subjects receive consistent confirmatory testing against a gold standard.	If negative tests are not confirmed against a gold standard, false-negative tests will not be recognized. If positive tests are not confirmed against a gold standard, false-positive tests will not be recognized. This may result in incorrect sensitivity and specificity calculations.
Spectrum bias	Patients with extremes of severity of injury or illness are enrolled, rather than patients with a representative range of severity.	A nonrepresentative sample may skew results, impairing external validity. It can affect sensitivity and specificity, because a test may readily diagnosis severe disease but not subtle disease.
Selection bias	A nonconsecutive sample of patients undergoes testing.	A nonrepresentative sample makes extrapolation of the result to other populations unreliable.
Manufacturer-specific variations in MRI protocols and sequences	Manufacturers may differ in the technical features of the equipment, which can be manipulated to obtain images.	Results cannot be readily reproduced outside of the research environment or using different equipment.

Biases associated with the application of a reliable and uniform diagnostic standard are one of the most common and serious problems in studies of diagnostic imaging. At its heart, any valid study of a new diagnostic test must compare the test result to some measure of “truth,” representing the actual presence or absence of disease in each patient. Without such a comparison, we have no reliable way to determine whether the new test under scrutiny has provided true-positive, false-positive, true-negative, or false-negative results. In a perfect scenario, an unequivocal diagnostic standard known as the gold standard (sometimes called the criterion or reference standard) exists and is applied in every case. For example, in a study of CT scan for appendicitis, if all patients (regardless of CT scan result) were to undergo appendectomy and surgical pathology were used to define the gold standard, a strong comparison could be made between CT interpretation and pathology. Even pathologists make mistakes, so gold standards are not themselves perfect, but this approximates “truth” as closely as we can reasonably expect in clinical medical research. Often in medicine, it is unreasonable or unethical for each research subject to undergo the gold standard test. For example, patients with normal CT scans might not be reasonably expected to undergo appendectomy. In these cases, a mixed diagnostic standard is often used, consisting of pathology reports (for patients with abnormal CT undergoing surgery) and clinical follow-up (in those with normal CT who are spared immediate appendectomy).

Several deviations from this ideal can occur, leading to bias. First, in some cases, it is unclear what the appropriate diagnostic standard should be. This is particularly a problem in diseases that rely on diagnostic imaging for diagnosis, with no confirmation possible by laboratory, microbiologic, pathologic, or surgical means. Consider a study of nondisplaced fractures, which are usually diagnosed by x-ray. If a new diagnostic test is compared with x-ray, what gold standard should be applied? Pathology is not performed to diagnose such fractures. Clinical follow-up could be applied as a diagnostic standard to determine which injuries heal without intervention and which require immobilization, but this may not accurately reflect the presence or absence of fracture. Some nondisplaced fractures might heal uneventfully without any specific treatment and might be miscategorized through follow-up as “nonfractures.” Other nonfracture injuries might be immobilized unnecessarily by practitioners because of continued patient symptoms and might be miscategorized as “fractures.” Still, clinical follow-up is sometimes the only pragmatic diagnostic standard and is arguably a more important outcome than the actual presence or absence of disease or injury. That said, researchers should be cautious in reporting their results in such cases as sensitivity and specificity, since these suggest a confirmed diagnosis. Instead, reporting of clinical outcomes in each diagnostic group (e.g., the percentage of patients healing uneventfully without immobilization in the MRI-negative group) is a more accurate description of the variables measured by such a study.

Often, researchers solve this problem of the lack of an independent diagnostic standard in a biased manner: they assume either the historical imaging test or the new imaging test to be correct in all cases. Whether the new or the old test is assumed to be correct, errors in interpretation are likely to occur. For example, if the new test is (in truth) superior to the historical test, it may discover disease when the historical test was negative or show no disease when the historical test was positive. If the historical test is considered the diagnostic standard, the new test results may be incorrectly categorized as false-positive and false-negative test results. As a result, the calculated sensitivity and specificity for the new test will be lower than is truly the case.

Often, researchers presume the new imaging test to be correct and assume discrepancies between results of the new test and the older test to reflect errors of the older test. By this definition, the new test cannot yield false-positive or false-negative results, and the sensitivity and specificity are by definition 100%. This flawed use of a new diagnostic test as its own gold standard is called incorporation bias. More subtly, incorporation bias may occur when the final clinical diagnosis is considered the diagnostic standard but the results of the imaging tests are available to the clinicians and thus influence the final clinical diagnosis. Blinding of clinicians to the results of the imaging tests under comparison can prevent this but is often not performed.²⁵ For example, imagine a study comparing CT and MRI for the diagnosis of hip fracture. If CT and MRI are compared with the final clinical diagnosis, it would appear that they are being appropriately subjected to an independent diagnostic standard. However, if the orthopedist rendering the final clinical diagnosis is aware of the imaging results and believes MRI to be more accurate, the clinical diagnosis will not be an independent standard but rather will reflect the MRI result. Many studies of MRI that we discuss later suffer from some degree of incorporation bias.

Another common error is verification bias (also called workup bias), which occurs when only some subjects undergo definitive confirmation with a diagnostic reference test. For example, patients with abnormal imaging results may undergo definitive testing with biopsy and pathologic analysis, whereas patients with normal imaging results may not. In effect, normal imaging results are assumed to be correct, whereas abnormal test results are subjected to further scrutiny against the gold standard. The results of such analyses are often incorrect. If some normal imaging results are actually falsely negative but are not discovered by definitive testing, the sensitivity of the imaging test may be overestimated. In the case of studies of MRI, this is a strong possibility because normal MRI is rarely followed by surgical intervention. Nishikawa et al.²⁶ estimated the role of verification bias in a study of MRI for meniscal tears of the knee. Using only patients undergoing arthroscopy following abnormal MRI, they calculated a sensitivity and specificity of 85% and 31%, respectively. After accounting for the possibility of undetected meniscal tears in patients with normal MRI, the estimates of sensitivity and specificity ranged from 29% to 95% and from 3% to 92%, respectively—demonstrating the potential importance of verification bias in misestimating test accuracy.²⁶

Incorporation and verification biases have particular importance in evaluation of studies of MRI, because the technologic sophistication and exquisitely detailed images of MRI may lead to the seductive conclusion that MRI is always correct or that “seeing is believing.” However, evidence suggests that many findings on MRI do not represent real or clinically important pathology. For proof of this, we turn to an area of MRI application where pathologic confirmation of MRI findings is possible, MR mammography for breast cancer screening. Based on pathology results, MRI has excellent sensitivity but poor specificity—with one false-positive test for every two true positives, according to a meta-analysis of 19 studies.²⁷ In the realm of musculoskeletal MRI, the poor specificity of MRI is demonstrated by studies of asymptomatic volunteers. Among subjects without back pain or neurologic symptoms, 25% to 50% have abnormal lumbar spine MRI.^28-29 Asymptomatic subjects undergoing wrist MRI have “abnormal” signal, simulating pathology.^30-32 Sugimoto et al.³¹ found that 50% of wrists in asymptomatic subjects had abnormal signal of the triangular fibrocartilage, making this MRI finding useless diagnostically.

Spectrum bias may influence the results of a study of diagnostic tests. Spectrum bias occurs when the population studied is skewed toward patients with either mild or severe disease or injury, rather than representing the full or typical range of clinical presentations. Although the results may be valid for similarly skewed populations, they may not reflect the performance of the test in a different population with a different distribution of disease severity. Many studies of MRI suffer from this bias because they examine populations of patients referred to specialty orthopedic clinics or undergoing operative interventions. These populations may be enriched in patients with disease or injury, compared with the typical population seen in an emergency department. The effect of spectrum bias on the apparent performance of a test can vary. For example, imagine a study of patients with massive pulmonary embolism and hemodynamic instability admitted to a medical intensive care unit. CT scan in such a population might appear to be 100% sensitive for detection of pulmonary embolism—which might well be true for large central pulmonary emboli in critically ill patients but might not be true for small, subsegmental pulmonary emboli in stable emergency department patients complaining only of chest pain or dyspnea. When evaluating a study of diagnostic tests, it is therefore essential to review the study setting and enrollment criteria, which may reveal spectrum bias.

A related phenomenon is selection bias, which can create a nonrepresentative subject sample at the point of study enrollment. Consecutive enrollment of all patients with a given complaint prevents selection bias, whereas a nonconsecutive sample is highly subject to this bias. Consider the example of a study of MRI in patients with negative hip x-ray after a fall. If all patients with negative hip x-ray are enrolled, selection bias is avoided. If, however, only those patients whom the investigator chooses to enroll become study subjects, the population may differ markedly from the larger population of all patients with falls and hip pain. Selection bias often results in spectrum bias and verification biases (described earlier)—the enrolled subjects may be sicker or more injured than the average patient, and performance of diagnostic tests may differ in this group.

Selection, spectrum, and verification biases can overlap and strongly affect the results of studies of disease or injury incidence and prevalence. Accurate studies of this type should be population-based, with consecutive enrollment of all patients meeting inclusion criteria. If, instead, only selected patients are enrolled and studied, the results will likely reflect the investigators’ biases. For example, if only those patients with a high clinical suspicion of injury are studied with definitive imaging after normal x-ray, the population of patients will likely be enriched with patients with injuries. The rate of detection of injuries with definitive imaging in this group will likely be higher than in the original pool of all patients, and the performance of x-ray will appear particularly poor because many injuries may be detected that were missed by x-ray. If patients with a lower pretest probability of disease had been studied, most negative x-rays would likely have been true negative, and the apparent performance of x-ray would be better. A study of this type may simply reflect the investigators’ desire to prove that x-ray is a poor test for fracture.

MRI is subject to additional problems of validation because of the proliferation of proprietary pulse sequences developed by different manufacturers. It can be difficult to reproduce the precise combination of magnetic field and radio frequency variations on another manufacturer’s equipment.⁷ In addition, when a study fails to demonstrate good diagnostic qualities for MRI, it may simply reflect a failure to select the best pulse sequence or imaging planes to demonstrate the pathology, rather than an across-the-board inability of MRI to make an accurate diagnosis of the condition.

Magnetic Resonance Imaging Diagnostic Accuracy for Epidural Abscess

Perhaps the most important musculoskeletal application of MRI in the emergency department is in the evaluation of potential spinal cord lesions (Figures 15-6 and 15-7; see also Chapter 3). Spine MRI constituted 29% of emergency department MRI in a study by Rankey et al.¹ Although trauma is likely the single leading indication, MRI is frequently performed for evaluation of cord compression or other abnormalities from metastatic disease, infectious processes, inflammatory disease (e.g., transverse myelitis), or demyelinating disease (e.g., multiple sclerosis). With some technical caveats related to diagnostic confirmation, as described earlier in our discussion of evidence-based medicine, MRI is generally considered the diagnostic standard for these conditions. MRI provides outstanding soft-tissue contrast, allowing extrinsic spinal cord compression, inflammatory lesions within the cord itself, and adjacent soft-tissue masses to be recognized. In its evidence-based appropriateness guidelines, the ACR gives MRI its highest appropriateness rating for investigation of myelopathy except in the case of trauma, where CT is rated as the initial test of choice because of its better depiction of bone.³³ Here, we consider some evidence for emergency department MRI for suspected spinal epidural abscess, comparing it with alternative imaging strategies.

Delay in diagnosis of spinal epidural abscess is associated with devastating neurologic outcome, suggesting that early diagnosis and treatment are particularly important. The diagnosis is not readily evident in many emergency department patients, mandating advanced imaging in patients with almost any level of clinical suspicion. Davis et al.³⁴ performed a case-control study of 63 emergency department patients with an ultimate diagnosis of spinal epidural abscess and found diagnostic delay (defined as multiple emergency department visits or admission without a diagnosis and greater than 24 hours from presentation until definitive imaging) in 75% of cases. Only 13% of patients presented with the classic triad of fever, spine pain, and neurologic abnormalities, demonstrating that imaging must be performed with less clinically overt presentations to avoid misdiagnosis. Outcomes were worse in patients with delayed diagnosis, with 45% having persistent motor weakness, compared with 13% of those with more rapid diagnosis (odds ratio = 5.65, 95% CI 1.15 to 27.71). Darouiche³⁵ reviewed the medical literature and concluded that the most important predictor of neurologic outcome in spinal epidural abscess is neurologic status beforesurgery—suggesting that diagnostic imaging before progression of neurologic signs and symptoms is essential to the improvement of outcomes. Because of the rarity of the condition, which occurs in only about 1 in 10,000 hospital admissions, validated clinical risk stratification guidelines do not exist; emergency physicians must rely on a high index of suspicion and low threshold for imaging whenever the diagnosis is suspected.

Methodologically rigorous trials of diagnostic imaging tests for spinal epidural abscess do not exist. Most published studies are retrospective case series, without the possibility of a control group or direct comparison of two diagnostic studies in the same patient. Because most studies rely on retrospective record review based on final diagnosis, the possibility of undiagnosed disease limits the validity of results. Patients with negative imaging tests are not generally included in these studies, and negative spine imaging studies are not verified against an independent diagnostic standard, such as clinical follow-up or surgical findings and cultures.

Despite the limited evidence, guidelines for spinal imaging have been promulgated by professional societies. The ACR (Table 15-5) ranks MRI of the spine without and with contrast as best (nine on a nine-point scale) in the infectious disease patient and seven of nine in the patient with painful myelopathy. CT earns a lower rating in the ACR criteria, between three and seven depending on the clinical scenario.³⁶ Unfortunately, these guidelines suggest imaging for the patient with an existing myelopathy (neurologic deficit localizing to the spinal cord). As described earlier, the presence of neurologic deficits is associated with poor outcomes, so emergency physicians should consider imaging for suspected spinal epidural abscess in the absence of neurologic abnormalities. For example, fever and back pain may be sufficient indication to perform neurologic imaging when an alternative diagnosis has not been identified.

Table 15-5 American College of Radiology Appropriateness Criteria: Infectious Disease Patient with Myelopathy

Procedure	ACR Rating	Comments
MRI spine without and with contrast	9
MRI spine without contrast	8
CT spine without contrast	6	If MRI unavailable or contraindicated
X-ray myelography	5	If MRI not feasible. Usually performed in conjunction with CT
CT spine with contrast	5
Myelography and postmyelography CT spine	5	Problem solving or if MRI unavailable or contraindicated
Nuclear medicine In-111 white blood cell scan of the spine	4	May be combined with bone scan to diagnose osteomyelitis
X-ray spine	3	To assess stability

From American College of Radiology: ACR Appropriateness Criteria.® Myelopathy. 2008. (Accessed at http://www.acr.org/SecondaryMainMenuCategories/quality_safety/app_criteria/pdf/ExpertPanelonNeurologicImaging/MyelopathyDoc8.aspx.) Reprinted with permission of the American College of Radiology, Reston, VA. No other representation of this material is authorized without expressed, written permission from the American College of Radiology. Refer to the ACR website at www.acr.org/ac for the most current and complete version of the ACR Appropriateness Criteria®.”Rating Scale: 1,2,3 Usually not appropriate; 4,5,6 May be appropriate; 7,8,9 Usually appropriate.Variant 6, Infectious disease patient. The full ACR criteria include other clinical variants with other imaging recommendations. Only the primary imaging recommendations (i.e., procedures rated as usually appropriate) and alternatives (i.e., procedure rated as may be appropriate) are included in this truncated list.

Magnetic Resonance Imaging for Occult Hip Fracture

Hip fractures, including fractures of the femoral head and neck, intertrochanteric region, and subtrochanteric region, can be radiographically occult when nondisplaced, particularly in elderly patients with significant osteopenia. If not recognized and treated with surgery or non-weight-bearing status, nondisplaced fractures can become displaced, increasing morbidity.^46-47 Rogers et al.⁴⁸ found an association between early surgical fixation of fractures (before 72 hours) and improved survival, decreased infectious complications, decreased length of stay, and decreased costs. However, although the authors attempted to control for differences between patients receiving early or later fixation, the study’s retrospective methods likely obscure a baseline difference in the underlying health of the subjects, which in turn likely influenced decisions about timing of interventions and may have strongly influenced outcomes. A Cochrane meta-analysis of randomized trials did not show a survival benefit of surgery compared with nonsurgical treatment, though the five included studies totaled only 428 patients.⁴⁷

Multiple imaging techniques can be applied when x-rays do not reveal a hip fracture but clinical suspicion remains high. Representative x-ray, MR, and CT images of radiographically occult hip fractures are shown in Figures 15-8 to 15-12.

The evidence comparing various imaging approaches is limited. The ACR recommends limited field-of-view MRI, Tc-99m bone scan with single photon emission computed tomography (SPECT), or repeat x-ray in 10-14 days as the most appropriate tests in a patient with normal x-ray but suspected insufficiency fracture because of osteoporosis or chronic steroid use, with a rating of nine of nine (most appropriate). Unfortunately, the ACR recommendation does not address the more common emergency department scenario of suspected acute traumatic fracture in the elderly patient with osteoporosis. The ACR grades CT scan as having an appropriateness of one—this disparity in “appropriateness” between MRI and CT does not appear justified by the quality of evidence, as we explain later. Delayed repeat x-rays are also of little use to the emergency physician attempting to determine patient disposition.

The ACR’s strong recommendation for MRI and against CT when hip fracture is suspected appears at odds with other imaging practices and evidence. In patients with severe multisystem trauma, CT scan is routinely used to evaluate for fracture of the cervical spine and pelvis and is believed to have excellent sensitivity. CT for these indications is reviewed in Chapters 3 and 13. MRI is not routinely performed to assess for fracture in that setting following normal CT. The multiplanar capability of CT and high spatial resolution (discussed earlier) are routinely accepted as sufficient to evaluate for cortical fracture. It seems counterintuitive that high-energy trauma should be cleared by CT while CT should be considered inadequate with ground-level, low-energy falls in the elderly, the common scenario for suspected hip fracture.

The clinical dilemma faced by the emergency physician caring for a patient with hip pain and no definite fracture by x-ray is common and real: admit the patient for further workup (difficult in a time of limited hospital beds and cost containment), hold the patient in the emergency department until MRI is available (potentially worsening the significant problem of emergency department crowding), perform alternative imaging in the emergency department, or discharge the patient with non-weight-bearing status until additional imaging is completed.

Let’s examine the evidence for MRI and other imaging strategies. High-quality studies directly comparing CT and MRI are limited: a PubMed search using the terms “MR CT hip fracture” with the limits “randomized controlled trial,” “clinical trial,” and “meta-analysis” finds one match. “MR hip fracture” with the same limits identifies three studies, two of which are not relevant to our clinical question.⁵¹ Our clinical question is really divided into several parts:

We examine each of these in turn.

What Is the Sensitivity of X-ray for Hip Fracture?

X-ray is likely highly sensitive for hip fracture, between 90% and 99%. Parker retrospectively reviewed 825 consecutive admissions for hip fracture; 822 of the patients underwent initial hip x-ray. The authors found 10 cases of “misinterpretation” of the initial x-ray and 3 cases in which the fracture was not recognized, even in retrospect with knowledge of the final diagnosis. If these cases are all considered false-negative results, x-ray identified 809 of 822 patients with fractures (98.4%).⁴⁶ Perron et al.⁵² reviewed the literature and found x-ray to be negative in only 1% to 9% of hip fractures. Dominguez et al.³ found that x-ray identified 219 of 243 (90%) patients with hip fractures. The additional 24 fractures were identified in 62 patients who underwent MRI during the emergency department visit. The authors performed structured follow-up 1 month after the initial evaluation in patients who did not undergo MRI and found no additional fractures. Use of this mixed diagnostic standard, although pragmatic, represents verification bias because not all patients with negative x-ray underwent MRI. Perhaps some patients with normal x-rays had no clinical fracture identified at follow-up but would have had MRI abnormalities consistent with fracture had they undergone initial MRI. Moreover, it is unclear whether all “fractures” identified by MRI were real; this assumes that all MRI abnormalities represent true-positive results, which is clearly not the case in other scenarios, as reviewed earlier in this chapter. Thus care should be taken in stating from this study that the sensitivity of x-ray is 90%—the true sensitivity might be higher or lower.

What Is the Sensitivity of Magnetic Resonance Imaging for Hip Fracture in Patients With Negative X-Ray?

MRI has been designated by the ACR⁴⁹ as the “most appropriate” imaging study when clinical suspicion of hip fracture persists despite negative x-ray. However, the methodologic quality of studies of MRI is surprisingly poor. Most studies of MRI lack an appropriate diagnostic standard, often suffering from incorporation bias by assuming MRI findings to be correct. The number of patients studied is small. No studies directly compare MRI and CT using an appropriate independent diagnostic standard, as we explain later.

What Is the Sensitivity of CT for Hip Fracture in Patients With Negative X-Ray?

CT is designated in ACR guidelines⁴⁹ as “least appropriate” (score of one out of nine). The ACR comments that axial imaging with CT may miss fractures in the axial plane and that reformatting of images is necessary—essentially ignoring that multiplanar, high-resolution CT reformations are now widely available. In the past, CT was thought to miss small impacted fractures and nondisplaced fractures parallel to the axial plane, but as described earlier, modern thin-slice CT with multiplanar reconstruction reduces the likelihood of missing such fractures (see Figure 15-5).⁵³

Can Limited Field-of-View Magnetic Resonance Imaging Provide Rapid and Accurate Diagnosis of Hip Fracture?

MRI with a limited field of view focused on the hip and using T1-weighted pulse sequences primarily for fracture detection can be performed in less than 15 minutes at lower cost than wider MR views of the hip and pelvis with greater numbers of pulse sequences. Small studies of this technique assert 100% sensitivity and specificity but suffer from serious methodologic flaws, including wide confidence intervals and selection, incorporation, and verification bias.

Quinn and McCarthy⁵⁴ prospectively evaluated 20 patients with suspected hip fracture and nondiagnostic x-rays, using a limited MR technique with a surface coil and a 7-minute protocol limited to a single coronal T1-weighted image. They report 100% sensitivity, but the validity of the final diagnosis is uncertain, given that the gold standard was the MRI and clinical outcome—a gross example of incorporation bias. Of the 13 patients diagnosed with fracture by this MR technique, 3 (23%) were treated nonsurgically; perhaps none of these patients had true fractures. The small number of patients in this study would result in broad confidence intervals for the reported sensitivity, even if the study findings are correct.

Rizzo et al.⁵⁵ reported on 62 consecutive patients with negative hip x-ray, further evaluated with both MRI and radionuclide bone scan. The MRI protocol was limited to T1-weighted images and required less than 15 minutes to complete. The authors concluded that this limited MRI technique was as accurate for diagnosis as bone scan, but no independent diagnostic standard was applied.

Lim et al.⁵⁶ reported on 57 patients with negative or equivocal hip x-ray who underwent limited MRI and found that a total of 56% of scans were positive for femoral neck fracture (14%), intertrochanteric fracture (9%), or other stable fractures (33%). As with the previous studies, however, MRI findings were assumed to be true positive, with no valid independent diagnostic standard.

Chana et al.⁵⁷ reported on a series of 35 patients undergoing limited T1-weighted MRI of the hip using an 8-minute imaging protocol following nondiagnostic x-ray. They concluded that MRI revealed pathology in 29 patients, including 21 femur fractures. However, no diagnostic standard was applied to confirm MRI findings.

Magnetic Resonance Imaging for Occult Scaphoid Fracture

Fractures of the scaphoid bone are an important and sometimes difficult diagnosis in the emergency department. The proximal scaphoid is supplied with blood primarily via branches of the radial artery that enter the distal scaphoid and perfuse the proximal pole in a retrograde fashion. Scaphoid fractures threaten this blood supply and can lead to avascular necrosis, fracture nonunion, and chronic arthritis. These outcomes may occur despite timely diagnosis and treatment; the risk for nonunion has been reported as 5% to 12% even with appropriate treatment.⁶⁷ However, when delay in diagnosis occurs, poor outcomes may be attributed to misdiagnosis. Missed scaphoid fractures are a frequent cause of litigation and payouts for medical malpractice,⁶⁸ accounting for 41% of claims involving the wrist, according to one source.⁶⁹ Most scaphoid fractures are likely seen with x-ray, but occult scaphoid fractures, clinically suspected but not seen on plain x-ray, remain a major concern in emergency medicine. Faced with the possibility of an unseen fracture, emergency physicians historically have immobilized patients in thumb spica splints, with outpatient imaging repeated to determine the presence of fracture. Because only 7% to 40% of suspected occult scaphoid fractures are ultimately confirmed, most patients with possible injury are subjected to needless immobilization and follow-up, which may be costly in terms of lost work productivity and additional medical expenses.^70-73 Immediate MRI in the emergency department has been proposed as an option to reduce diagnostic delay and unnecessary immobilization. Early or emergency department MRI might detect not only occult scaphoid fractures but also other clinically important injuries, such as nonscaphoid fractures of the wrist and forearm and soft-tissue injuries requiring operative repair. Early detection might lead to earlier appropriate therapy and better outcomes. Conversely, normal MRI might rule out these injuries, allowing discontinuation of immobilization, earlier return to work, and a decreased need for follow-up.

In this section, we consider the diagnostic options for suspected occult scaphoid fracture, with a focus on the evidence for MRI. We compare MRI with the alternative management schemes, including immobilization until repeat x-ray in 10 to 14 days, nuclear scintigraphy (bone scan), and immediate CT scan. X-ray, MR, and CT images of acute wrist injuries are shown in Figures 15-13 to 15-16.

For all imaging modalities, determination of sensitivity and specificity is complicated by the uncertainty of the appropriate diagnostic standard. For example, as in the cases we considered earlier in this chapter, it is unclear whether MRI signal abnormalities always indicate true-positive results. When x-ray, CT, or nuclear scintigraphy results are discrepant with MRI findings, it is uncertain which modality reflects the truth. A researcher’s choice in selecting the gold standard often reflects biases rather than even-handed consideration of the scientific evidence.

In its appropriateness criteria, the ACR recommends casting with follow-up x-ray in 10 to 14 days or MRI as the most appropriate imaging modalities for suspected acute scaphoid fracture following normal plain x-rays. Both approaches are given a score of eight on the one-to-nine scale used by the ACR, with nine being most appropriate.⁷⁴ The ACR credits CT an appropriateness score of four unless MRI or casting and follow-up are contraindicated.⁷⁴ This discrepancy in appropriateness scores might be taken to imply that strong evidence shows the superiority of MRI to CT. However, a careful review of the evidence considered by the ACR and other published work shows this is not the case, as we explain later. The limited evidence supporting the superiority of one diagnostic protocol over others is reflected in the marked practice variation shown in one international survey of 105 hospitals. The usual diagnostic examination after normal x-rays was MRI in 31 (30%), CT in 19 (18%), and scintigraphy in 14 (13%) of 105 examinations. Some hospitals used no formal protocol, leaving the choice of modality to the individual practitioner. Others used a combination of modalities, including MRI to confirm negative CT findings.⁷⁵

As we did for MRI of the hip, let’s consider several basic questions to place in context the value of MRI in diagnosis of occult scaphoid fracture:

How Sensitive Are Initial X-rays in the Diagnosis of Scaphoid Fracture?

As we discussed earlier in this chapter, a high-quality study of any diagnostic test requires the consistent application of a reliable diagnostic standard in all patients. In the case of x-ray and scaphoid fracture, the gold standard is not clear, and most studies also fall short of this ideal by evaluating only a small percentage of patients with negative x-rays with additional imaging. This leads to verification bias (described earlier), with potential effects on the apparent sensitivity of x-ray. First, because many patients with negative x-ray undergo no diagnostic confirmation, some of these negative x-rays may be false negatives. If all patients with negative x-rays who undergo no confirmatory testing are assumed to have true-negative results, the sensitivity of x-ray may be wrongly inflated. Second, because only a select few high-risk patients may be subjected to advanced imaging techniques, selection and spectrum bias are likely to occur, with a large number of patients found to have injuries despite negative initial x-ray. When injuries are detected in this group, this may suggest that the sensitivity of x-ray was very low, if a similar rate of injury is assumed in all patients with negative initial x-rays. However, the selection- and spectrum-biased group undergoing follow-up imaging likely does not represent the rate of injury in the original population. Relying on the rate of injury in this group results in underestimation of the sensitivity of x-ray in the larger population.

With this said, what is the sensitivity of initial x-ray, and what is the rate of scaphoid fracture identified in patients with negative initial x-ray? The sensitivity of x-ray is variously reported as 59% to 79%, and 7% to 40% of patients with normal initial x-rays are shown to have scaphoid fracture with some form of follow-up.^76-77 This wide range undoubtedly is related to the variable diagnostic standards applied in various studies and to the selection and spectrum biases described earlier. In studies that include only patients who follow up because of continued pain, the likelihood of fracture is inflated. Asymptomatic patients likely do not have fractures but are often excluded from these studies.

Are Follow-up X-rays Sensitive for Initially Radiographically Occult Scaphoid Fracture?

Determining the sensitivity of follow-up x-ray is complicated by the same issues of diagnostic standard that thwart studies of initial x-ray. An independent standard is needed, and it is unclear whether MRI findings, follow-up x-ray, or clinical outcome should be accepted as proof of fracture. Not surprisingly, authors variably selected the diagnostic standard.

Breitenseher et al.⁷⁸ retrospectively studied 42 patients with negative initial plain x-rays but clinical suspicion of scaphoid fracture, using 6-week follow-up x-rays as the diagnostic standard. MRI at 7 days after injury revealed an apparent fracture in 33%. Compared with follow-up x-ray, MRI was nearly 100% sensitive and specific. Several problems with this study limit its application to emergency department patients. First, the study is small and involves a biased sample, not a consecutive emergency department sample of patients with wrist trauma. Second, MRI performed at 7 days after injury may differ significantly from MRI performed in the emergency department. Perhaps early emergency department MRI would show more signal abnormalities, actually representing edema but not fracture, and would therefore be too nonspecific to be clinically useful. In addition, MRI performed too early in the course of injury evolution would possibly be less sensitive, if signal abnormalities representing fracture take hours to appear.

Gabler et al. prospectively enrolled 121 consecutive patients with wrist injury and negative initial x-rays, performing MRI and clinical follow-up in all patients. This study lacks a single diagnostic standard; in some cases the final clinical diagnosis is considered “correct,” whereas in other cases MRI appears to be the gold standard. The authors reported that 28 occult scaphoid fractures were identified, a rate of 23% in patients with nondiagnostic initial x-ray. The authors noted that MRI diagnosed additional occult injuries; including 15 other fractures of carpal bones; 26 other bone injuries, including distal radius, ulna, and metacarpal fractures; and “bone bruises”—MRI signal abnormalities of uncertain clinical importance. Some patients with bone bruises and one with a reported triquetrum fracture diagnosed by MRI were asymptomatic at follow-up and had no further treatment—again, calling into question the importance or clinical validity of MRI. MRI also revealed other soft-tissue injuries including cartilaginous and ligamentous injuries. The authors concluded that only 30% of the detected scaphoid fractures were truly initially occult because 70% were recognized on review of the initial x-rays by “experienced surgeons”—but this reflects a retrospective review of images without blinding to the presence of continued symptoms and likely exaggerates the diagnostic accuracy of x-ray. Several additional methodologic concerns limit application of this study to the emergency department population. We cannot determine the sensitivity of initial x-ray from this study because the number of patients with positive initial x-rays is not reported. Moreover, the rate of scaphoid injury in this population may not be similar to that in U.S. emergency departments because this study examined patients evaluated by a trauma service “fracture clinic” in Austria.⁷⁶ MRI in this study was performed an average of 3 days after injury, limiting our ability to assess its sensitivity and specificity if performed immediately in the emergency department. The authors suggested that clinical examination and repeat radiography are sufficient in most cases, although their protocol was resource intensive, with three follow-up examinations at 10, 24, and 38 days by hand surgeons and prolonged immobilization of patients. Oddly, the study abstract reports randomization of patients, though the study protocol described in the manuscript does not involve randomization.⁷⁶

Low and Raby compared 50 sets of initial (negative) and follow-up x-rays with paired MRI as diagnostic standard, using a retrospective sample of 50 patients. Four expert readers reviewed the follow-up x-rays for scaphoid fracture, and the sensitivity and specificity of the readers varied widely, from 9% to 49% and 80% to 93%, respectively. Not surprisingly, given these broad variations in accuracy among readers, the interobserver reliability proved to be poor, with a coefficient of 33%. An interobserver reliability coefficient of 60% is generally considered the threshold for a “reliable” diagnostic test. The authors concluded that x-rays are too insensitive, nonspecific, and unreliable to be used as a method for confirming or ruling out scaphoid fracture. However, it is unclear whether patients with MRI abnormalities that were not detected on follow-up x-ray would be harmed by misdiagnosis. The authors did not report the treatment or outcome in patients with MRI abnormalities, and it is possible that most did not require any specific intervention. If patients with negative follow-up x-rays have good clinical outcomes regardless of MRI appearance, then x-ray may be diagnostically sufficient. In addition, the patient population sampled was not representative of an emergency department population because only those (presumably high risk) patients who underwent both follow-up x-ray and MRI were included.⁷⁹

Are Follow-up X-rays Adequate to Assess for Scaphoid Fracture in Children Before Skeletal Maturity? Do These Patients Require Magnetic Resonance Imaging or Other Advanced Imaging?

Given the limited evidence for follow-up x-rays in adults (described earlier), it should come as little surprise that controversy remains about the utility of x-ray in children. X-ray should be less sensitive in children with skeletal immaturity because bone density and mineralization are incomplete. The age at which ossification of the scaphoid and other carpal bones occurs varies. The scaphoid begins ossification around 5 to 6 years of age and completes ossification around 13 to 15 years. Before completion of ossification, the scaphoid may be nearly invisible on x-ray.⁸⁰ Fortunately, pediatric scaphoid fractures are thought to be uncommon, although it is possible that this simply reflects underdiagnosis.⁸¹ Large methodologically strong studies in children are lacking. Cook et al.⁸² examined 18 pediatric patients (8 to 15 years old) undergoing wrist MRI within 2 days of injury. MRI identified six fractures and four cases of bone marrow edema. The authors asserted that marrow edema did not represent fracture and that MRI was 100% sensitive—though the same questions about diagnostic standard exist. The role of MRI in pediatric evaluation remains uncertain. Immobilization and clinical follow-up remain viable options.

How Sensitive Are CT and Magnetic Resonance Imaging for Scaphoid Fracture and Other Wrist Injuries?

The ACR^82a rates CT as four of nine in its appropriateness for evaluation of suspected scaphoid fracture following normal x-rays, below both MRI and casting with repeat x-ray in 10 to 14 days. However, CT scan has several potential advantages in the emergency department. Unlike MRI, CT is widely available at all times of day in many emergency departments. Wrist CT for assessment of fracture does not require contrast administration. The radiation exposure from extremity CT is of relatively little clinical concern because the extremities are generally less radiosensitive than organs such as the breast, thyroid, and gonads. Pregnant patients can undergo extremity CT with essentially no radiation exposure to the fetus.⁷⁴ However, CT scan does not image ligaments and cartilage structures of the wrist that may require surgical repair, a disadvantage relative to MRI—but also of repeated x-rays. Let’s examine some studies evaluating the diagnostic performance of CT and MRI.

Kusano et al.⁸³ reported MRI findings in 52 consecutive patients with suspected scaphoid fractures—though no clear diagnostic standard was defined. MRI findings were interpreted as indicative of fracture in 18 of 52 patients (33%). In 16 of the 18 patients with abnormal MRI, CT scan was also performed and demonstrated 16 fractures. Because CT was not performed in patients with normal MRI, it is uncertain whether it would have revealed injuries missed by MRI—a case of verification bias. The clinical significance of the imaging findings on both CT and MRI in these cases is uncertain.

Brydie and Raby⁷³ studied 195 patients undergoing MRI within 14 days for occult scaphoid fracture after negative x-ray. Based on MRI, 19% of patients had scaphoid fractures, 14% had distal radius fractures, and 5% had other carpal bone fractures. According to a survey of the treating clinicians, management was altered in 92% of patients—though it is unclear whether clinicians were required to state their anticipated management before MRI results were available. Moreover, it is not clear that changes in management benefited patients. For example, MRI may have shown signal abnormalities, prompting prolonged immobilization, when no cortical fracture was actually present. No independent diagnostic standard was established in this study, so the accuracy of MRI findings is unknown.

Kumar et al.⁸⁴ prospectively enrolled patients with clinically suspected scaphoid fracture but negative x-ray, performing MRI within 24 hours. In the 22 patients enrolled, MRI diagnosed six scaphoid fractures, two distal radius fractures, and a hamate fracture. Of the enrolled patients, 13 had no fracture discovered by MRI. The authors concluded that early MRI prevents unnecessary immobilization and is “sensitive” for occult scaphoid fracture, but this study is quite limited. An independent diagnostic standard is not defined in this study; MRI abnormalities are assumed to be real, and negative MRI is assumed to be correct. This incorporation bias is a serious design flaw that limits the study validity.²⁴ In addition, selection bias may play an important part in this study because consecutive patients were not enrolled. Finally, small studies such as this with no reported adverse outcomes should include confidence intervals rather than reporting a “0%” misdiagnosis rate. Confidence intervals can be estimated by the 3/n methodology when n > or = 30 subjects have been enrolled.⁸⁵ In this case, if no adverse events occur in n subjects, the upper 95% confidence interval limit is 3/n. For example, if no adverse events are seen in 30 subjects, the upper 95% confidence interval limit is 3/30, or 10%. A larger sample, 99 subjects, gives a narrower limit of 3/99, or 3%. If 300 subjects are enrolled but no adverse events occur, the likely limit of adverse events falls to 3/300, or 1%. Naturally, in a small study such as the Kumar et al. study described here, the confidence intervals are wide, and we cannot be certain that MRI would safely identify all patients with fractures in a larger study.

Groves et al.⁸⁶ reported a case of scaphoid fracture detected with follow-up nuclear scintigraphy, MRI, and 6-week follow-up x-ray but missed by 16-detector CT using 0.75-mm slice thickness and 0.5-mm multiplanar reconstruction. Many problems limit this report. The MR, nuclear medicine, and x-ray images were not interpreted with blinding to one another, so the interpretation of one set of images was influenced by the others. The patient underwent casting because of the imaging abnormalities, but as usual we cannot know with certainty whether a clinically important fracture was present. Case reports suffer from many limitations in general, representing the lowest echelon of medical evidence. Still, the possibility that CT could miss a real scaphoid fracture should be considered. The authors concluded that CT may miss fractures when neither cortical nor trabecular bone displacement are present, though it is unclear how common or clinically important this scenario may be.

Karle et al.⁸⁷ reviewed 30 patients with scaphoid nonunion and noted that 70% had been seen by a physician and either misdiagnosed or treated nonoperatively without success. They suggest that all suspected scaphoid fractures undergo CT imaging to guide treatment decisions, with nondisplaced waist fractures treated with casting. Proximal fractures, and comminuted or displaced fractures seen on CT would undergo operative therapy by their classification. However, these recommendations have not been validated by any study of outcomes or compared with management based on x-ray or MRI findings.

Lozano-Calderon et al.⁸⁸ compared the performance of x-ray alone, CT alone, or the combination of x-ray and CT in determining the presence of scaphoid fracture displacement. Six observers reviewed imaging studies from 30 cases of “known” fracture, 10 displaced and 20 nondisplaced. Intra and interobserver reliability was higher for CT or CT with x-ray compared with x-ray alone. The authors reported the sensitivity and specificity of the three categories of imaging, but the diagnostic standard is unclear, making the validity of these values uncertain. This study includes patients with fractures visible on x-ray, so it is impossible to extrapolate to patients with occult scaphoid fracture.

Memarsadeghi et al.⁷¹ prospectively compared CT scan and MRI with a gold standard of follow-up x-ray at 6 weeks in 29 patients with suspected scaphoid fracture but normal initial x-rays. CT and MRI were performed within 6 days of trauma, again limiting the applicability of results to the performance of CT or MRI in the emergency department. CT was performed using narrow collimation (0.5 mm, the acquired slice thickness) and thin multiplanar reconstructions (0.7 mm, the thickness of the digitally reconstructed slices). MRI was performed with a 1-tesla machine using STIR, spin–echo T1, and gradient–echo T2 pulse sequences. Follow-up x-rays demonstrated eight cortical fractures and three trabecular fractures in 11 patients (38%). No false-positive results were obtained with CT or MRI. CT identified all cortical fractures but failed to detect trabecular fractures. Early MRI detected all patients with fractures but did not demonstrate five of eight cortical fractures. The authors thus reported the sensitivity of CT for cortical fractures as 100% but the sensitivity of MRI as only 38%. This small study suggests a complementary role for CT and MRI, with CT being superior for cortical injury and MRI for trabecular injury. However, the clinical importance of fractures solely involving the trabeculae is uncertain. Perhaps detection of cortical fractures is clinically sufficient, in which case CT may be adequate. Given the wide confidence intervals resulting from this small study, which overlap for CT and MRI, it is impossible to say with certainty that one modality is superior to the other. Further study is required.

Cruickshank et al.⁸⁹ prospectively evaluated 47 nonconsecutive emergency department patients with suspected scaphoid fracture but negative x-ray. Patients were immobilized in a plaster cast and underwent CT within 24 hours. The diagnostic standard was the clinical diagnosis at 10 days, including information from clinical exam, x-ray, and MRI—which was not performed in all patients. The study thus suffers from verification bias. The authors asserted that CT identified 17 of 18 carpal fractures (94.4% sensitive and 100% specific) and detected all scaphoid fractures. One patient with a trapezium fracture detected by CT had a second fracture of the capitate, detected only by MRI. Whether this abnormality represented a real fracture and whether patient management would be altered by the additional diagnosis remain in question. This small study requires validation in a consecutive, unbiased sample, with a more consistent diagnostic standard and a larger sample size to narrow confidence intervals.

Nguyen et al.⁹⁰ retrospectively reviewed CT scan results from 118 patients with suspected occult scaphoid fracture visiting a fracture clinic in follow-up after an initial emergency department visit. CT (performed within 1 week of injury in 70% of patients) revealed scaphoid fractures in 22% of patients, as well as capitate, triquetral, and radial fractures in an additional 10%. The authors concluded that CT in this setting has a high diagnostic yield. However, the population of patients may differ substantially from those seen in an emergency department because it represents a high-risk group who were not only referred for follow-up but carried through, possibly because of continued symptoms. Ubiquitous use of CT in the emergency department for patients with negative wrist x-ray might have a much lower diagnostic yield. This study has no diagnostic standard; all imaging abnormalities demonstrated by CT are assumed to represent real and clinical important fractures.

Although CT appears to be a promising modality, a well-designed, large, prospective emergency department study with consecutive enrollment, uniform application of a diagnostic standard, and evaluation of clinical outcomes has not been performed.

How Sensitive Is Radionuclide Scintigraphy (Bone Scan) for Scaphoid Fracture?

Radionuclide scintigraphy is a long-standing technique to assess for scaphoid and other occult fractures. In the presence of injury, increased blood flow and radionuclide signal occurs. The clinical importance of such increased signal is not clear. For example, bruising of bone without unstable cortical fracture might result in a similar appearance to clinically significant cortical injury. It is not certain that management decisions should be based on scintigraphy findings, although normal scintigraphy likely excludes injury.

Groves et al.⁹¹ studied 51 patients undergoing both 16-detector multislice CT and 99Tcm methylene diphosphonate scintigraphy on the same day. The investigators quantified the radionuclide uptake in all patients and compared uptake in patients with fractures seen on CT and those with no fracture seen on CT. They found that 16 patients had apparent fracture on CT and matching scintigraphic abnormalities. Another 7 patients had abnormal scintigraphy but normal CT. In these patients, the mean scintigraphic activity was approximately half of that in patients with abnormal CT, suggesting that a threshold could be identified to discriminate cortical fractures from bone bruising or other clinically unimportant injuries. Patients with no abnormalities on CT but with scintigraphic abnormalities had no fractures identified on follow-up MRI or radiographs. Although intriguing, these findings have not been validated in a larger study, and the clinical relevance is uncertain. Though a reliable threshold to discriminate cortical fractures from bruising would improve the specificity of nuclear scintigraphy, it might remain more practical to perform CT. If CT is considered the diagnostic standard in this study, then 7 of 35 patients with normal CT had a false-positive nuclear scintigraphy study—a 20% rate that could lead to unnecessary immobilization, follow-up, additional imaging, or even surgery.

A review in the Best Evidence Topic series compared scintigraphy to MRI for the diagnosis of occult scaphoid fractures.⁹² Using a Medline search from 1966 until March 2005, 11 studies were identified, with only 4 studies relevant to the question, “In an adult with a clinically suspected scaphoid fracture that is not evident on plain x-ray, is magnetic resonance imaging better than bone scintigraphy at reaching a diagnosis?” Only 145 patients had been enrolled in studies directly comparing the two modalities, with variable diagnostic standards. Based on this limited evidence, MRI appears to be slightly more accurate and can anatomically localize soft-tissue injuries, guiding surgical planning. MRI is also faster than bone scan.⁹² Nonetheless, this review highlights the need for large, multicenter trials with appropriate power, well-defined diagnostic standards, and relevant clinical outcomes. Additional small studies with poor methodology are unlikely to provide reliable data comparing scintigraphy to other modalities.

Is Magnetic Resonance Imaging Cost-Effective for Evaluation of Suspected Scaphoid Fracture?

Our discussion has highlighted the limited evidence for MRI in scaphoid fracture. Uncertainties remain about the clinical significance of abnormalities on MRI, which could in some cases represent false-positive results and prompt unnecessary treatment. Given these limitations, can we determine whether emergent MRI is a cost-effective strategy?

Cost-effectiveness estimates typically include direct costs of medical imaging and indirect costs including follow-up medical diagnostic testing and treatments. Moreover, nonmedical economic costs such as lost productivity because of unnecessary immobilization must be considered, though these costs can be difficult to estimate accurately. Cost-effectiveness of a diagnostic strategy may vary widely depending on disease or injury prevalence, the sensitivity and specificity of the diagnostic tests used, and the costs of over treatment and undertreating. For example, early imaging might be cost-effective in a working population, where lost wages might be a significant cost associated with delayed diagnosis and unnecessary prolonged immobilization. However, in a population of high school students or retired workers, this cost might be negligible. Moreover, if MRI were performed in every patient with wrist trauma and negative x-rays, rather than in a select few with a high rate of injury, the cost of care might be increased. The costs associated with MRI may vary tremendously depending on the specificity of the test, which, as we saw earlier, remains in question. MRI enthusiasts recommend early MRI to allow early detection of injury (when MRI is positive) and to limit unnecessary immobilization (when MRI is negative). However, this assumes high sensitivity and specificity of MRI. If MRI results in frequent false-positive findings (e.g., signal abnormalities that indicate edema but not clinically important fracture), early MRI might actually increase the rate of immobilization and requirement for follow-up. Most existing studies of MRI sensitivity and specificity examine MRI after 1 or more days of delay after injury. It is possible that very early MRI could perform quite differently, for example, becoming nonspecific by identifying some degree of edema in all patients with wrist trauma and pain. With these issues in mind, let’s consider some studies of cost-effectiveness of MRI.

Raby⁹³ used low-field MRI with a dedicated extremity coil in 56 patients within 4 days of injury and 53 patients with persistent symptoms at 10 days to 6 weeks. Seven scaphoid fractures were identified in the early MRI group, with 6 radius and 4 other fractures. In the late MRI group, 14 scaphoid, 9 radial, and 3 other fractures were identified. The author’s cost analysis suggests lower total costs with early, rather than later, scanning. This study assumes all MRI findings to be correct. If extended to an emergency department population, it is unclear whether the proportion of positive and negative imaging studies would be similar, and costs might differ markedly as a result.

Saxena et al.⁹⁴ compared the cost of usual care (immobilization, clinical follow-up, and repeated x-rays) with early MRI, finding an advantage for early MRI. Gooding et al.⁹⁵ performed a similar estimate, finding that early MRI was less expensive (NZ$470 vs. NZ$533). Both of these models are limited by numerous assumptions about direct and indirect costs, test sensitivity, and injury prevalence.

Brooks et al.⁹⁶ randomized 28 patients with suspected but radiographically occult scaphoid fracture to standard care (immobilization, clinical follow-up, and repeat x-rays), with or without MRI. In an accompanying cost-effectiveness analysis, the MRI strategy was slightly more expensive ($594) than usual care ($428), though not statistically different. The small study size means that a real difference in costs might have existed but failed to be detected because of lack of statistical power, with only 17 patients in the control group and 11 in the MRI group. The cost per day saved by MRI by preventing unnecessary immobilization was $44—although such cost savings might vary wildly in a different population because this depends on the patient’s income, job requirements, and ability to perform productively despite immobilization. Clearly differences in these costs would occur between a world-class concert violinist and a retired or unemployed worker. Moreover, this study examined MRI within 5 days of injury; an immediate MRI in the emergency department strategy might result in more savings by eliminating more days of immobilization. The costs of the two strategies would also depend on the prevalence of fractures; fewer days of immobilization would be saved if most patients were found to have fractures.

Jenkins et al.⁷⁰ estimated mean costs of five diagnostic strategies for occult scaphoid fracture in the United Kingdom in 2006, assuming best-published diagnostic performance of each test: immobilization and 2-week follow-up (£144), MRI (£302), radionuclide bone scan (£243), CT (£202), and ultrasound (£113). The investigators also estimated costs using worst-published diagnostic performance, with similar results. Although the investigators attempted to account for variations in the reported sensitivity and specificity of each imaging strategy in the medical literature, this may not reflect the real-world performance of any of these tests. Moreover, the reported sensitivity and specificity of each modality are largely based on flawed studies such as those we reviewed earlier in this chapter. The cost estimates in this model also assume that a “negative” advanced imaging study would terminate the workup and associated medical costs, which may not be true in a real patient with persistent symptoms. This study also does not incorporate nonmedical costs of immobilization and lost productivity, which would be expected to increase the cost of the follow-up strategy compared with advanced imaging strategies.

Current studies of diagnostic imaging for occult scaphoid fracture suffer from significant methodologic flaws. MRI is likely to be sensitive for scaphoid fracture, other carpal and forearm fractures, and other soft-tissue injuries. The specificity of MRI remains a potential problem, with little evidence yet that MRI abnormalities mandate treatment or alter patient outcomes. Most patients with equivocal x-rays in the emergency department do not require immediate MRI, although in some cases, this may be cost-effective. If “definitive” imaging is required emergently but MRI is unavailable or contraindicated, noncontrast CT or nuclear medicine bone scan may be reasonable alternatives. These modalities do not provide the soft-tissue contrast or resolution required to identify important soft-tissue injuries such as triangular fibrocartilage tears. CT arthrography can be performed but is generally not an emergency department imaging test.