Preliminary evaluation of dual-energy CT to quantitatively assess bone marrow edema in patients with diabetic foot ulcers and suspected osteomyelitis

All patients being treated at the diabetic foot expert center of our tertiary referral hospital who had undergone a DECT scan of the feet between January 2018 and January 2021 were retrospectively selected and screened. Inclusion criteria were clinically suspected osteomyelitis and an open ulcer at the time of scanning. Patients were excluded in the case of concomitant foot diseases (e.g., active Charcot osteoarthropathy or recent trauma) on the foot with the ulcer of interest and if the patients’ digital files were unavailable. Patients were divided into two clinical study groups: osteomyelitis and no-osteomyelitis depending on the final clinical diagnosis. This diagnosis was based on clinical data and findings during follow-up including depth of the ulcer (i.e., bone contact), fever, laboratory values indicating inflammation, MRI findings, bone biopsy findings, ulcer progression or healing, amputation, and structural changes on the weighted average or mixed images of the CT scans (CT was not used to evaluate BME for clinical diagnosis). The Medical Ethics Review Committee provided a waiver for this study.

All DECT scans were performed using a dual-source CT scanner (SOMATOM Force; Siemens Healthcare) using 80 kVp (tube A) and 150 kVp with Sn filtration (tube B). Automatic attenuation-based tube current modulation was applied on both tubes with a reference mAs of 150 for tube A and a reference mAs of 380 for tube B. Other CT parameters were a Qr54d or Qr40d kernel (medium smooth), rotation time of 0.5 s, collimation of 0.6 mm, slice thickness of 1.5 mm, increment of 1.5 mm, mean CT dose index volume (CTDIvol) of 6.0 mGy (range 4.1–6.6 mGy), and mean dose length product (DLP) of 152.8 mGy/cm (range 97.0–206.4 mGy/cm).

After each DECT acquisition, three datasets were reconstructed: an 80-kVp dataset, an Sn150-kVp dataset, and a weighted average or mixed dataset from both tubes. The weighted average was used to simulate and replace a conventional CT scan and was used in clinical practice. The scans were analyzed using the BME application in SyngoVia post-processing software (SyngoVia VB40; Siemens Healthcare) with a threshold of 100 HU and a maximum of 800 HU. Using a three-material decomposition algorithm, yellow bone marrow, red bone marrow, and bone mineral could be differentiated, allowing for subtraction of calcium and the creation of VNCa images. The default settings for color-coded maps of bone marrow with densities between − 150 HU and 100 HU were used.

All scans were independently assessed by two investigators (M.A.M. and A.G.) who were blinded for the radiology report. The ulcer was identified on the weighted average DECT dataset using clinical information related to ulcer location. In difficult cases, an experienced musculoskeletal radiologist (M.M.) assisted in locating the ulcer. A circular region of interest (ROI) was manually placed in sagittal or coronal VNCa images within bones with suspected osteomyelitis as close as possible to the index ulcer (Figs. 1 and 2). If multiple ulcers with suspected osteomyelitis were present, all ulcers were assessed. If the bone of interest was too small to place a circular ROI, a freehand ROI was drawn as large as possible without including irrelevant structures. On the same location in the contralateral foot, a reference ROI with the same surface area as the ROI in the bone of interest was drawn. If this location was not suitable as a reference (e.g., due to amputation, Charcot, ulceration, or logistics), the talus of the contralateral or (if the contralateral talus was also not suitable) the ipsilateral talus was used as a reference. The CT values of the ROI in the affected bone and the reference bone were measured. If CT values differed more than 50 HU between observers, an experienced musculoskeletal radiologist (M.M.) performed a third measurement. This measurement was used in further calculations, except for calculations related to observer variability. Measurements with less than 50 HU difference between observers were reviewed by two radiologists (M.M. and N.L.W.) with, respectively, 30 years and 5 years of experience in musculoskeletal radiology. Additionally, they independently and qualitatively assessed the cases for presence of BME. In cases where minimally one radiologist considered BME to be present, a positive finding of BME was used for further analysis. BME has CT values of approximately 0 and will be displayed as green on the color-coded map. Higher CT values are more yellow or eventually red and lower CT values are blue or purple (Figs. 1 and 2).

Descriptive statistics were used to describe differences in demographics between groups. Data regarding CT values were tested for normality using the Kolmogorov–Smirnov test. An independent-samples t-test was used to test for significance between the osteomyelitis and no-osteomyelitis group regarding CT values on both the affected and reference side and a chi-squared test was used to test for significance in the qualitative analysis. The difference between the affected side and reference side was tested for significance using a paired-samples t-test. A p < 0.05 was considered statistically significant. Inter-observer agreement was determined by calculating the intraclass correlation coefficient (ICC) or Cohen’s kappa. An ICC < 0.5 was considered “poor”; 0.5–0.74, “moderate”’ 0.75–0, “good”; and > 0.9, “excellent” [25]. A kappa < 0 was considered “poor”; 0–0.20, “slight”; 0.21–0.40, “fair”; 0.41–0.60, “moderate”; 0.61–0.80, “substantial”; and > 0.81, “almost perfect” [26]. Additionally, Bland–Altman plots with limits of agreement were obtained. Using receiver operating characteristic (ROC) analysis and Youden’s index, a cutoff CT value for determining BME was proposed. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated. All statistical tests were performed using SPSS (version 26.0, released 2019, IBM SPSS Statistics for Windows; IBM Corporation).

Osteomyelitis was clinically diagnosed in 23 of 56 ulcers. There was an amputation in 9 cases and 2 patients died before the ulcer was closed in the osteomyelitis group. Of the remaining 12 ulcers in this group, the mean time until closing was 18.5 ± 14.7 weeks. There was no clinical indication of osteomyelitis in 33 of the 56 ulcers. Amputation was performed in 10 cases before the ulcer was closed in the no-osteomyelitis group and there were no deaths before ulcers closed. The remaining 23 ulcers in this group had a mean time until closing of 16.3 ± 20.3 weeks.

There was a significant difference in CT values of the bone marrow in the bone adjacent to the ulcer between the osteomyelitis and no-osteomyelitis groups (− 17.23 ± 34.96 HU and − 69.34 ± 49.40 HU, p < 0.001) (Fig. 3). The CT values of the reference location did not differ significantly between the two groups (− 68.90 ± 33.09 HU and − 86.26 ± 32.00 HU, p = 0.054) (Fig. 3). Regarding the osteomyelitis group, there was a significant difference between the ulcer location and reference location (p < 0.001). In the no-osteomyelitis group, the ulcer location showed similar CT values to the reference location (p = 0.052). The inter-observer agreement was “good” for measurements on the ulcer location (ICC = 0.858) and “moderate” for the reference measurements (ICC = 0.675). Bland–Altman plots were made to represent the difference between observers on the ulcer side (Fig. 4) and reference side (Fig. 5). There were 6 measurements with a difference between the observers of more than 50 HU of which 2 on the ulcer side. These 6 measurements were repeated by the third observer (M.M.). The original measurements (by M.A.M. and A.G.) are presented in the Bland–Altman plots (Figs. 4 and 5). The radiologists agreed with 96% of the other measurements. In the remaining 4%, the radiologists would have placed the ROI further from the cortex and they would have excluded one case due to metal artifacts. However, changing the ROI placement would not have influenced the results. On the ulcer side, there was a bias of − 1.69 with an SD of 28.5. The 95% agreement limits were − 57.6 and 54.3. On the reference side, the bias was 1.30 with an SD of 33.4 and 95% agreement limits of − 64.2 and 66.8. The area under the ROC curve was 0.821 (Fig. 6). With a cutoff value of − 40.1 HU, sensitivity was 87.0%, specificity was 72.7%, PPV was 69.0%, and NPV was 88.9%.

BME was marked as present in 37 of the 56 cases (66.1%). There was a significant difference between the osteomyelitis and no-osteomyelitis groups in BME presence (p = 0.006). There was disagreement in 2 cases and Cohen’s kappa was “almost perfect” (κ = 0.89). Sensitivity was 87.0%, specificity was 48.5%, PPV was 54.1%, and NPV was 84.2%.