Nondisplaced fractures on hip CT: added value of dual-energy CT virtual non-calcium imaging for detection of bone marrow edema using visual and quantitative analyses

Abstract

Background

Dual-energy computed tomography (DECT) virtual non-calcium (VNCa) imaging can demonstrate bone marrow edema.

Purpose

To evaluate the added value of DECT VNCa imaging for detection of non-displaced fractures on hip CT and to determine the attenuation value cut-off to identify bone marrow edema around fractures.

Material and Methods

We identified 35 patients who underwent DECT due to suspicion of a non-displaced fracture of the proximal femur or pelvic bone. Twenty-seven non-displaced fractures were present in 24 patients, while 11 patients had no fractures. Radiological, clinical follow-up, or surgical record was the standard of reference. Two radiologists visually interpreted CT images in two sessions: one with bone reconstruction images only and one with bone reconstruction images and VNCa images. Diagnostic performances were compared between the two reading sessions. Quantitative analyses of CT numbers on VNCa images were performed and a cut-off value was obtained through receiver operating characteristic (ROC) analysis.

Results

In the visual analysis, area under the curve, sensitivity, and negative predictive value were improved in the second session with VNCa images for both readers. In the quantitative analysis, a cut-off value of –55.3 HU yielded 100% sensitivity, 94% specificity, 95.4% accuracy, 69.0% positive predictive value, and 100% negative predictive value for detection of non-displaced fractures.

Conclusion

Visual and quantitative analyses of VNCa images may be useful in detection of non-displaced fractures in the proximal femur and in pelvic bones.

Keywords

Dual-energy computed tomography CT bone marrow edema occult hip fracture virtual non-calcium

Introduction

Proximal femoral fractures are a major cause of morbidity and mortality in the elderly (1). Delayed diagnosis of proximal femoral fractures can result in fracture displacements and higher incidence of complications, leading to increased morbidity (2,3). Although non-displaced pelvic bone fractures are generally treated conservatively, they may cause prolonged pain and impair hip function if not properly treated (4). Early and accurate diagnosis of proximal femur and pelvic bone fractures is therefore important to direct appropriate management and improve patient outcomes.

Computed tomography (CT) is often the first cross-sectional study performed when a hip fracture is suspected, with reported detection rates of radiographically occult fracture in the range of 83–100% (5 –8). CT may fail to demonstrate subtle non-displaced fractures (9). Traumatic bone marrow edema (BME) is sensitively visualized by magnetic resonance imaging (MRI) in radiographically occult fractures. Although MRI is the reference standard for assessing BME, it is less accessible in an emergency setting, slower to obtain, and may be contraindicated for certain patients (10). As an alternative modality, dual-energy CT (DECT) has recently been shown to be informative regarding BME.

Several studies have evaluated the ability of DECT virtual non-calcium (VNCa) images to detect BME at various parts of the body, including two studies regarding visual assessment of proximal femur fracture (9,11–14). To our knowledge, this is the first study to perform quantitative analysis of VNCa imaging for diagnosis of non-displaced fractures at the proximal femur and pelvic bone. We hypothesized that VNCa imaging could assist in detection of occult fractures on hip CT by both visual and quantitative analyses of BME.

The purpose of this study was to evaluate the added value of VNCa imaging for detection of non-displaced fractures on hip CT and to determine the CT number cut-off to identify BME around fractures.

Material and Methods

Study population

This study was approved by the institutional review board of our hospital and the requirement for informed consent was waived due to its retrospective nature. A total of 160 hip CT studies performed with dual-source CT systems between 1 May 2017 and 31 March 2018, were retrospectively reviewed. Of these studies, those without a history of trauma were excluded (n = 7). All displaced hip fractures were excluded (n = 76) because such fractures do not represent diagnostic challenges for conventional CT. Fractures at the sacrum were excluded (n = 3) because the sacrum is not imaged in its entirety by hip CT at our hospital. Patients with hip prostheses were excluded (n = 39). After exclusions, 35 hip CT examinations were included in our study (Fig. 1).

Fig. 1.

Flow chart of the patient selection process.

Non-displaced fractures at the proximal femur and pelvic bones, including pubis and acetabulum, were included. The diagnoses of two patients with fractures were determined by follow-up CT that showed displacement. Two patients were confirmed as negative for fracture by subsequent hip MRI or bone scan. For the remaining 31 patients, we used surgical reports or clinical follow-up during the 30 days after presentation as our standard of reference based on previous studies (9,11). Patients were considered positive for fracture if they had a surgical report describing repair of a hip fracture (n = 13) or a consultation letter by an orthopedic surgeon stating the decision for non-surgical fracture management (n = 9). Patients who did not have a surgical report describing fracture repair or a consultation letter from an orthopedic surgeon describing non-surgical fracture management within 30 days of clinical follow-up were classified as being negative for fracture (n = 9).

Image acquisition

CT scans were obtained with a dual-source CT system (Somatom Definition Flash; Siemens, Erlangen, Germany). Tube voltages were set at 100 kVp and 140 kVp with an activated tin filter. DECT acquisition was performed using a detector configuration of 32 × 0.6 mm, pitch of 0.6, rotation time of 0.5 s, and effective milliampere second value of 160 mAs with automated attenuation-based tube current modulation. Combined raw projection data from both tubes were used to generate regular CT images reconstructed with a section width and increment of 1 mm using a standard bone deconvolution kernel (B75f; Siemens Healthcare). Dual-energy datasets were reconstructed using a section width of 1.5 mm and an increment of 1 mm with a standard dual-energy deconvolution kernel (D34; Siemens Healthcare). The dual-energy datasets were then post-processed using a VNCa software algorithm (syngo.via, version VB30B; Siemens Healthcare). Coronal and axial VNCa series were generated using default parameters: relative contrast ratio of 1.44; minimum and maximum CT values of 100 and 800, respectively; smoothing filter range of 4; section width of 1 mm; and increment of 2 mm. The resulting non-calcium images were color-coded (bone marrow setting in Syngo Dual Energy) with a shift from blue/purple representing fat to green representing water and further to yellow for increasing red marrow/blood content. The overlay images of VNCa and standard CT images were used for image analysis.

Image interpretation

CT images were visually evaluated on a Picture Archiving and Communication System (PACS) workstation (M-View, Marotech, Seoul, Republic of Korea). For visual analysis, two radiologists (reader 1 with three years of clinical radiology experience; reader 2 with 14 years of musculoskeletal radiology experience) independently reviewed the CT images. The readers did not participate in study population selection and were blinded to clinical information. Image review consisted of two separate sessions with a four-week interval. In the first session, the readers evaluated standard bone reconstruction images. In the second session, the readers evaluated bone reconstruction images with the corresponding VNCa BME images for the same cases. In both sessions, the cases were interpreted as being positive or negative for acute fracture at the proximal femur or pelvic bone, and the fracture sites were recorded.

Two radiologists (reader 3 with two years of musculoskeletal radiology experience; reader 4 with six years of clinical radiology experience) independently performed quantitative analyses of multiplanar reformatted images using a syngo.via viewer (version VB30B; Siemens Healthcare). Although these readers were blinded to the results of readers 1 and 2, they were not blinded to the clinical information or reference standard and were able to review bone reconstruction images during measurement. CT numbers were obtained using circular regions of interest (ROIs) from the locations of highest edema intensity at fracture sites. The ROIs were also placed at normal bone marrow of nonfracture sites in both femur necks, both intertrochanteric areas, both pubic bones, and both acetabula in all patients with or without fractures. The ROIs were placed on either coronal or axial images. The areas of the ROIs were set in the range of 0.4–0.5 cm².

Statistical analysis

Fisher’s exact tests and Mann–Whitney U-tests were used to compare the male-to-female ratio and age between individuals with and without fractures. An exact McNemar’s test and weighted generalized score test were used to compare diagnostic performance between interpretations under the two sessions. DeLong’s method was utilized for analysis of the area under the curve (AUC).

Two sample t-tests were used to compare the CT values between fracture sites with BME and non-fracture sites. Receiver operating characteristic (ROC) curve analysis was used to evaluate the CT numbers and determine the quantitative cut-off value of the CT numbers. The CT number that maximizes the Youden’s index was chosen as the cut-off value. Based on this cut-off value, we calculated sensitivity, specificity, accuracy, positive predictive value (PPV), and negative predictive value (NPV) (15).

Cohen’s kappa (κ) statistics and intraclass correlation coefficient were used to determine the levels of inter-observer agreement in visual and quantitative analyses, respectively. κ values were assigned as follows: < 0.20, poor; 0.21–0.40, fair; 0.41–0.60, moderate; 0.61–0.80, good; and > 0.81, excellent (16). R software, version 3.4.1 (R Foundation for Statistical Computing, Vienna, Austria; http://www.R-project.org) was used for statistical analysis. Differences with P < 0.05 were considered statistically significant.

Results

Demographic characteristics

Our study included 35 patients (25 men, 10 women; mean age = 78.3 years; age range = 60–94 years). Twenty-four patients (17 men, 7 women; mean age = 80.2 years; age range = 66–94 years) had 27 non-displaced fractures at the proximal femur or pelvic bone; 11 patients (eight men, three women; mean age = 74.0 years; age range = 60–87 years) did not have fractures. There was no significant difference in male-to-female ratio or age between groups.

Twenty-seven non-displaced fractures were included. The fractures were observed at the femoral neck (n = 5), greater trochanter (n = 5), intertrochanteric area (n = 10), pubis (n = 5), and acetabulum (n = 2).

Evaluation of the added value of VNCa imaging in detecting non-displaced fractures

Table 1 shows the diagnostic predictive values for both readers in each reading session. The AUC, sensitivity, and NPV for detection of non-displaced fractures improved for both readers in the second session when VNCa images were included. AUC improved from 0.825 to 0.926 and from 0.963 to 1, sensitivity by 11% and 7%, and NPV by 14% and 15% for readers 1 and 2, respectively. Inter-observer agreement was good for both sessions, with improved kappa values during the second session (from 0.673 to 0.769).

Table 1.

Comparison of diagnostic performance between bone reconstruction images alone and bone reconstruction images with virtual non-calcium images in diagnosis of nondisplaced fractures.

Type of imaging and statistics	Reader 1	Reader 2
Bone reconstruction image
AUC	0.825 (0.702–0.948)	0.963 (0.912–1)
Accuracy (%)	78.9 (62.7–90.4)	94.7 (82.3–99.4)
Sensitivity (%)	74.1 (53.7–88.9)	92.6 (75.7–99.1)
Specificity (%)	90.9 (58.7–99.8)	100 (71.5–100)
PPV (%)	95.2 (76.2–99.9)	100 (86.3–100)
NPV (%)	58.8 (32.9–81.6)	84.6 (54.6–98.1)
Bone reconstruction image with VNCa images
AUC	0.926 (0.858–0.994)	1 (1–1)
Accuracy (%)	89.5 (75.2–97.1)	100 (90.7–100)
Sensitivity (%)	85.2 (66.3–95.8)	100 (87.2–100)
Specificity (%)	100 (71.5–100)	100 (71.5–100)
PPV (%)	100 (85.2–100)	100 (87.2–100)
NPV (%)	73.3 (44.9–92.2)	100 (71.5–100)
P value for pairwise comparison of ROC curves	0.066	0.149
P values calculated using McNemar’s test
Sensitivity	0.250	0.500
Specificity	1	1
P values calculated using weighted generalized score test
PPV	0.285	1
NPV	0.051	0.156

Numbers in parentheses are 95% confidence intervals.

VNCa, virtual non-calcium; AUC_, area under the ROC curve; PPV, positive predictive value; NPV, negative predictive value; ROC, receiver operating characteristic.

Additional interpretations of VNCa images enabled the two readers to correct several diagnostic errors made on the review of bone reconstruction images only (reader 1: four cases, reader 2: two cases; Table 2). Two non-displaced fractures were missed during the first session by both readers. One case was corrected by additional review with VNCa by both readers (Fig. 2), while the other case was corrected by reader 2 only (Fig. 3). Of the five fractures missed during the first session by only reader 1, two were correctly diagnosed during the second session. One false-positive finding by reader 1 was corrected by additional review of VNCa images, which demonstrated normal bone marrow (Fig. 4).

Table 2.

Number of false-negative or false-positive cases before and after virtual non-calcium image interpretation.

	Before virtual non-calcium image interpretation	After virtual non-calcium image interpretation
Reader 1
False-negative cases (n)	7	4
False-positive cases (n)	1	0
Reader 2
False-negative cases (n)	2	0
False-positive cases (n)	0	0
Both reader 1 and reader 2
False-negative cases (n)	2	1 (reader 1) 0 (reader 2)
False-positive cases (n)	0	0

Fig. 2.

A 73-year-old man who experienced a fall and subsequent left hip pain. (a) Coronal hip CT bone reconstruction image shows a subtle radiolucent line at the intertrochanteric area. Both readers interpreted the image as negative for fracture in the first session. (b) Coronal color-coded virtual non-calcium overlay image shows bone marrow edema around the subtle radiolucent line, which was correctly diagnosed as a non-displaced fracture by both readers.

Fig. 3.

An 82-year-old woman with slip down and left hip pain. (a, b) Axial hip CT bone reconstruction images show focal cortical buckling at the left anterior acetabular wall (a, short arrow) and left pubic bone (b, long arrow). Although both readers detected the left pubic bone fracture, both missed the left acetabular fracture (c, d). Axial color-coded virtual non-calcium overlay image shows bone marrow edema around cortical buckling at the left acetabulum and pubic bone, which facilitated the detection of non-displaced fractures. The left acetabular fracture was correctly diagnosed after review of the virtual non-calcium overlay image by reader 2.

Fig. 4.

An 86-year-old man with slip down and right hip pain. (a) Axial hip CT bone reconstruction image shows a subtle radiolucent line at the left femur neck. Reader 1, who was blinded to the clinical history, interpreted the image as a non-displaced left femur neck fracture. After additional review of virtual non-calcium overlay images (b), reader 1 corrected the diagnosis as negative for fracture with normal bone marrow at the suspected area.

Quantitative analysis

The inter-reader agreement of ROI-based CT numbers was excellent (0.949). The mean CT numbers for both readers were used for statistical analysis. Significant differences in CT numbers on VNCa images were found between fracture sites and non-fracture sites (P < 0.0001). We divided all CT measurements into two subgroups: the proximal femur group (femoral neck, intertrochanteric area) and the pelvic bone group (pubis, acetabulum). There were significant differences between fracture sites and non-fracture sites for each group (P < 0.0001) (Table 3).

Table 3.

CT numbers of the proximal femur and pelvic bone on virtual non-calcium images.

	Mean CT number	Range
Both proximal femur and pelvic bone
Fracture site	–4.9 ± 27.8	–55.3–49.4
Non-fracture site	–96.3 ± 23.9	–147.2– –24.4
P value	<0.0001
Proximal femur
Fracture site	4.4 ± 24.7	–51.4–49.4
Non-fracture site	–110 ± 19	–147.2– –55.2
P value	<0.0001
Pelvic bone
Fracture site	–33.9 ± 13.3	–55.3 – –13.6
Non-fracture site	–84.4 ± 21.3	–144.7 – –24.4
P value	<0.0001

Data are mean CT numbers and standard deviation from virtual non-calcium images.

ROC analysis of CT numbers at both the proximal femur and pelvic bone revealed an AUC of 0.996; a cut-off value of –55.3 HU for detection of BME provided overall sensitivity and NPV of 100%, specificity of 94.9%, accuracy of 95.4%, and PPV of 69.0% for detection of non-displaced fractures. ROC analysis of CT numbers at the proximal femur revealed an AUC of 1; a cut-off value of –53.3 HU for detection of BME accomplished 100% sensitivity, specificity, accuracy, PPV, and NPV for detection of non-displaced fractures. ROC analysis of CT numbers at the pelvic bone revealed an AUC of 0.981; a cut-off value of –55.3 HU for detection of BME produced 100% sensitivity and NPV, 91.1% specificity, 91.5% accuracy, and 36.8% PPV for detection of non-displaced fractures (Table 4).

Table 4.

Diagnostic performance of CT number cutoff values on virtual non-calcium imaging for identification of bone marrow edema in proximal femur and pelvic bone fractures.

	Both proximal femur and pelvic bone	Proximal femur	Pelvic bone
Cut-off value (HU)	–55.3	–53.3	–55.3
AUC	0.996 (0.991–1)	1 (1–1)	0.981 (0.950–1)
Accuracy (%)	95.4 (92.2–97.5)	100 (97.4–100)	91.5 (85.7–95.6)
Sensitivity (%)	100 (88.1–100)	100 (84.6–100)	100 (59.0–100)
Specificity (%)	94.9 (91.4–97.2)	100 (96.9–100)	91.1 (85.0–95.3)
PPV (%)	69.0 (52.9–82.4)	100 (84.6–100)	36.8 (16.3–61.6)
NPV (%)	100 (98.5–100)	100 (96.9–100)	100 (97.0–100)

Numbers in parentheses are 95% confidence intervals.

HU, Hounsfield unit; AUC_, area under the ROC curve; PPV, positive predictive value; NPV, negative predictive value; ROC, receiver operating characteristic.

Discussion

This study demonstrated increased AUC, sensitivity, and NPV for both readers when VNCa images were added to bone reconstruction images, with improved inter-observer agreement. The CT numbers were analyzed quantitatively, and a cut-off value of –55.3 HU achieved excellent accuracy, sensitivity, specificity, and NPV for detecting BME.

The AUC improved from 0.825 to 0.926 and from 0.963 to 1, sensitivity by 11% and 7%, and NPV by 14% and 15% for readers 1 and 2, respectively. The diagnostic performance of reader 2 during the second session was comparable with the results of the study by Kellock et al. (9). It is notable that the increase in diagnostic performance after adding VNCa images was greater in reader 1 (the less experienced reader) than reader 2, with lower P values for comparison of AUC and NPV. In addition, the inter-observer agreement improved from a kappa value of 0.673 to 0.769 after adding VNCa images. This suggests a potential role for the use of additional VNCa images to help less experienced radiologists or clinicians in emergency departments diagnose non-displaced fractures more sensitively and confidently. Similarly, Kaup et al. also demonstrated that less experienced radiologists particularly benefit from addition of VNCa images, and that inconclusive diagnoses of acute vertebral compression fractures were significantly decreased (17).

The four non-displaced fracture cases (three cases in reader 1 and two cases in reader 2, including one case that both readers missed) that were missed on bone reconstruction images were identified with the addition of VNCa images. In CT studies of patients with severe osteopenia, it is difficult to be certain that negative findings for fracture are truly negative. The increase in NPV achieved by adding VNCa images indicates the potential role of DECT as a diagnostic tool for exclusion of non-displaced fractures. There was one false-positive case in the first session, which was corrected during the second session upon detection of normal bone marrow around the suspicious radiolucent line. As such, VNCa images can also be helpful when there are ambiguous radiolucent lines, such as nutrient foramina, on bone reconstruction images.

The cut-off value of –55.3 HU in this study allowed detection of BME at the proximal femur and pelvic bone with high sensitivity, specificity, and NPV. Although no previous study has performed quantitative analyses for the proximal femur and pelvic bone region, the cut-off value we determined lies between those calculated by Petritsch et al. (14) (–47 HU) and Wang et al. (18) (–80 HU), both of which addressed BME at recent vertebral compression fractures. Ali et al. found that a cut-off value of 5.90 HU allowed detection of BME at the carpal bones with high sensitivity and specificity (13). Guggenberger et al. (19) identified the cut-off values of BME at the ankle mortise, talar dome, and talar body/head to be –52 HU, –70 HU, and –35 HU, respectively. These variances can be explained by differences in scanner hardware, kilovoltage settings, bones, and age groups examined in the studies. Future research with larger samples and variable age groups for the proximal femur and pelvic bone would help yield clinically usable cut-off values.

The sensitivity, specificity, accuracy, and NPV derived from ROC analysis were excellent. However, PPV was 69.0%, which was substantially lower than that of Kellock et al. (96–100%) and of visual analysis during the second session of our study (9). Among 14 false-positive cases with higher attenuation values than the cut-off value, two BME sites were related to severe degenerative changes of the hip joint. Although there are many causes of BME, including infection, osteonecrosis, or arthritis, interpretation in conjunction with bone reconstruction images can minimize false-positive results, as shown in the visual analysis of this and a previous study (9). At the remaining 12 locations with false-positive attenuation values, which comprised 10 acetabulum and two pubic bones, no causative lesions for BME were found. When divided into subgroups of the proximal femur and the pelvic bone, a cut-off value of –55.3 HU at the pelvic bone provided a lower PPV of 36.8%. Although the explanation for the high CT numbers at the acetabulum and pubic bone remains unclear, we speculated that the high attenuated area is related to red marrow composition, considering that the pelvic bone and spine are the last locations of fatty conversion and the first locations at which red marrow reconversion takes place (20).

There are several limitations in our study. First, a relatively small number of cases was included, especially the number of pelvic bone fractures, partly due to strict inclusion criteria. We excluded all displaced fractures and patients without trauma histories and focused on traumatic, non-displaced fractures. Second, we did not use MRI as a reference modality, given the limited availability of MRI in the emergency setting and cost. Because only one patient without fracture had undergone MRI, the BME noted on VNCa images could not be correlated with MRI. Third, the sensitivity and specificity values provided by ROC analysis are potentially biased because we did not perform objective measurements in an independent validation set. Finally, ROI analysis may take time to be adopted routinely in daily practice. However, using quantitative measurement in selected ambiguous cases could be a helpful and efficient option.

In conclusion, visual and quantitative analyses of VNCa imaging can depict BME around non-displaced fractures at the proximal femur and pelvic bone. Careful evaluations of VNCa images may reduce the need for additional imaging studies and achieve earlier diagnoses of non-displaced fractures.

Footnotes

Acknowledgements

The author(s) thank Young Lee for assistance with statistical analysis.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received the following financial support for the research, authorship, and/or publication of this article: This study was supported by a VHS Medical Center Research Grant, Republic of Korea (Grant No. VHSMC 18032).

References

Johnell

Kanis

JA.

An estimate of the worldwide prevalence and disability associated with osteoporotic fractures.

Osteoporos Int 2006; 17:1726–1733.

Peleg

Rozenfeld

Radomislensky

, et al. Policy encouraging earlier hip fracture surgery can decrease the long-term mortality of elderly patients. Injury 2014; 45:1085–1090.

Chana

Noorani

Ashwood

, et al. The role of MRI in the diagnosis of proximal femoral fractures in the elderly. Injury 2006; 37:185–189.

Oka

Monu

JU.

Prevalence and patterns of occult hip fractures and mimics revealed by MRI.

Am J Roentgenol 2004; 182:283–288.

Thomas

Williams

Carpenter

, et al. The validity of investigating occult hip fractures using multidetector CT. Br J Radiol 2016; 89:20150250.

Hakkarinen

Banh

Hendey

GW.

Magnetic resonance imaging identifies occult hip fractures missed by 64-slice computed tomography.

J Emerg Med 2012; 43:303–307.

Haubro

Stougaard

Torfing

, et al. Sensitivity and specificity of CT- and MRI-scanning in evaluation of occult fracture of the proximal femur. Injury 2015; 46:1557–1561.

Collin

Geijer

Gothlin

JH.

Computed tomography compared to magnetic resonance imaging in occult or suspect hip fractures. A retrospective study in 44 patients.

Eur Radiol 2016; 26:3932–3938.

Kellock

Nicolaou

Kim

SSY

, et al. Detection of bone marrow edema in nondisplaced hip fractures: Utility of a virtual noncalcium dual-energy ct application. Radiology 2017; 284:798–805.

10.

Mandell

Weaver

Khurana

Computed tomography for occult fractures of the proximal femur, pelvis, and sacrum in clinical practice: Single institution, dual-site experience.

Emerg Radiol 2018; 25:265–273.

11.

Reddy

McLaughlin

Mallinson

, et al. Detection of occult, undisplaced hip fractures with a dual-energy CT algorithm targeted to detection of bone marrow edema. Emerg Radiol 2015; 22:25–29.

12.

Song

Meta-analysis of dual-energy computed tomography virtual non-calcium imaging to detect bone marrow edema.

Eur J Radiol 2017; 95:124–129.

13.

Ali

Wong

Liang

, et al. Clinical utility of dual-energy CT analysis of bone marrow edema in acute wrist fractures. Am J Roentgenol 2018; 210:842–847.

14.

Petritsch

Kosmala

Weng

, et al. Vertebral compression fractures: Third-generation dual-energy CT for detection of bone marrow edema at visual and quantitative analyses. Radiology 2017; 284:161–168.

15.

Metz

CE.

Basic principles of ROC analysis.

Semin Nucl Med 1978; 8:283–298.

16.

Landis

Koch

GG.

The measurement of observer agreement for categorical data.

Biometrics 1977; 33:159–174.

17.

Kaup

Wichmann

Scholtz

, et al. Dual-energy CT-based display of bone marrow edema in osteoporotic vertebral compression fractures: Impact on diagnostic accuracy of radiologists with varying levels of experience in correlation to MR imaging. Radiology 2016; 280:510–519.

18.

Wang

Tsai

Chuang

, et al. Bone marrow edema in vertebral compression fractures: Detection with dual-energy CT. Radiology 2013; 269:525–533.

19.

Guggenberger

Gnannt

Hodler

, et al. Diagnostic performance of dual-energy CT for the detection of traumatic bone marrow lesions in the ankle: Comparison with MR imaging. Radiology 2012; 264:164–173.

20.

Malkiewicz

Dziedzic

Bone marrow reconversion - imaging of physiological changes in bone marrow.

Pol J Radiol 2012; 77:45–50.