Deep Learning to Differentiate Benign and Malignant Vertebral Fractures at Multidetector CT.

Background Differentiating between benign and malignant vertebral fractures poses diagnostic challenges. Purpose To investigate the reliability of CT-based deep learning models to differentiate between benign and malignant vertebral fractures. Materials and Methods CT scans acquired in patients with benign or malignant vertebral fractures from June 2005 to December 2022 at two university hospitals were retrospectively identified based on a composite reference standard that included histopathologic and radiologic information. An internal test set was randomly selected, and an external test set was obtained from an additional hospital. Models used a three-dimensional U-Net encoder-classifier architecture and applied data augmentation during training. Performance was evaluated using the area under the receiver operating characteristic curve (AUC) and compared with that of two residents and one fellowship-trained radiologist using the DeLong test. Results The training set included 381 patients (mean age, 69.9 years ± 11.4 [SD]; 193 male) with 1307 vertebrae (378 benign fractures, 447 malignant fractures, 482 malignant lesions). Internal and external test sets included 86 (mean age, 66.9 years ± 12; 45 male) and 65 (mean age, 68.8 years ± 12.5; 39 female) patients, respectively. The better-performing model of two training approaches achieved AUCs of 0.85 (95% CI: 0.77, 0.92) in the internal and 0.75 (95% CI: 0.64, 0.85) in the external test sets. Including an uncertainty category further improved performance to AUCs of 0.91 (95% CI: 0.83, 0.97) in the internal test set and 0.76 (95% CI: 0.64, 0.88) in the external test set. The AUC values of residents were lower than that of the best-performing model in the internal test set (AUC, 0.69 [95% CI: 0.59, 0.78] and 0.71 [95% CI: 0.61, 0.80]) and external test set (AUC, 0.70 [95% CI: 0.58, 0.80] and 0.71 [95% CI: 0.60, 0.82]), with significant differences only for the internal test set (P < .001). The AUCs of the fellowship-trained radiologist were similar to those of the best-performing model (internal test set, 0.86 [95% CI: 0.78, 0.93; P = .39]; external test set, 0.71 [95% CI: 0.60, 0.82; P = .46]). Conclusion Developed models showed a high discriminatory power to differentiate between benign and malignant vertebral fractures, surpassing or matching the performance of radiology residents and matching that of a fellowship-trained radiologist. © RSNA, 2024 See also the editorial by Booz and D'Angelo in this issue.

Comparing CT-Like Images Based on Ultra-Short Echo Time and Gradient Echo T1-Weighted MRI Sequences for the Assessment of Vertebral Disorders Using Histology and True CT as the Reference Standard.

BACKGROUND: Several magnetic resonance (MR) techniques have been suggested for radiation-free imaging of osseous structures. PURPOSE: To compare the diagnostic value of ultra-short echo time and gradient echo T1-weighted MRI for the assessment of vertebral pathologies using histology and computed tomography (CT) as the reference standard. STUDY TYPE: Prospective. SUBJECTS: Fifty-nine lumbar vertebral bodies harvested from 20 human cadavers (donor age 73 ± 13 years; 9 male). FIELD STRENGTH/SEQUENCE: Ultra-short echo time sequence optimized for both bone (UTEb) and cartilage (UTEc) imaging and 3D T1-weighted gradient-echo sequence (T1GRE) at 3 T; susceptibility-weighted imaging (SWI) gradient echo sequence at 1.5 T. CT was performed on a dual-layer dual-energy CT scanner using a routine clinical protocol. ASSESSMENT: Histopathology and conventional CT were acquired as standard of reference. Semi-quantitative and quantitative morphological features of degenerative changes of the spines were evaluated by four radiologists independently on CT and MR images independently and blinded to all other information. Features assessed were osteophytes, endplate sclerosis, visualization of cartilaginous endplate, facet joint degeneration, presence of Schmorl's nodes, and vertebral dimensions. Vertebral disorders were assessed by a pathologist on histology. STATISTICAL TESTS: Agreement between T1GRE, SWI, UTEc, and UTEb sequences and CT imaging and histology as standard of reference were assessed using Fleiss' κ and intra-class correlation coefficients, respectively. RESULTS: For the morphological assessment of osteophytes and endplate sclerosis, the overall agreement between SWI, T1GRE, UTEb, and UTEc with the reference standard (histology combined with CT) was moderate to almost perfect for all readers (osteophytes: SWI, κ range: 0.68-0.76; T1GRE: 0.92-1.00; UTEb: 0.92-1.00; UTEc: 0.77-0.85; sclerosis: SWI, κ range: 0.60-0.70; T1GRE: 0.77-0.82; UTEb: 0.81-0.92; UTEc: 0.61-0.71). For the visualization of the cartilaginous endplate, UTEc showed the overall best agreement with the reference standard (histology) for all readers (κ range: 0.85-0.93). DATA CONCLUSIONS: Morphological assessment of vertebral pathologies was feasible and accurate using the MR-based bone imaging sequences compared to CT and histopathology. T1GRE showed the overall best performance for osseous changes and UTEc for the visualization of the cartilaginous endplate. LEVEL OF EVIDENCE: 1 TECHNICAL EFFICACY: Stage 2.

Longitudinal MR-based proton-density fat fraction (PDFF) and T2* for the assessment of associations between bone marrow changes and myelotoxic chemotherapy.

OBJECTIVES: MR imaging-based proton density fat fraction (PDFF) and T2* imaging has shown to be useful for the evaluation of degenerative changes in the spine. Therefore, the aim of this study was to investigate the influence of myelotoxic chemotherapy on the PDFF and T2* of the thoracolumbar spine in comparison to changes in bone mineral density (BMD). METHODS: In this study, 19 patients were included who had received myelotoxic chemotherapy (MC) and had received a MR imaging scan of the thoracolumbar vertebrates before and after the MC. Every patient was matched for age, sex, and time between the MRI scans to two controls without MC. All patients underwent 3-T MR imaging including the thoracolumbar spine comprising chemical shift encoding-based water-fat imaging to extract PDFF and T2* maps. Moreover, trabecular BMD values were determined before and after chemotherapy. Longitudinal changes in PDFF and T2* were evaluated and compared to changes in BMD. RESULTS: Absolute mean differences of PDFF values between scans before and after MC were at 8.7% (p = 0.01) and at -0.5% (p = 0.57) in the control group, resulting in significantly higher changes in PDFF in patients with MC (p = 0.008). BMD and T2* values neither showed significant changes in patients with nor in those without myelotoxic chemotherapy (p = 0.15 and p = 0.47). There was an inverse, yet non-significant correlation between changes in PDFF and BMD found in patients with myelotoxic chemotherapy (r = -0.41, p = 0.12). CONCLUSION: Therefore, PDFF could be a useful non-invasive biomarker in order to detect changes in the bone marrow in patients receiving myelotoxic therapy. CLINICAL RELEVANCE STATEMENT: Using PDFF as a non-invasive biomarker for early bone marrow changes in oncologic patients undergoing myelotoxic treatment may help enable more targeted countermeasures at commencing states of bone marrow degradation and reduce risks of possible fragility fractures. KEY POINTS: Quantifying changes in bone marrow fat fraction, as well as T2* caused by myelotoxic pharmaceuticals using proton density fat fraction, is feasible. Proton density fat fraction could potentially be established as a non-invasive biomarker for early bone marrow changes in oncologic patients undergoing myelotoxic treatment.

MR imaging after patellar MACI and MPFL reconstruction: a comparison of isolated versus combined procedures.

OBJECTIVE: To qualitatively and quantitatively evaluate the 2.5-year MRI outcome after Matrix-associated autologous chondrocyte implantation (MACI) at the patella, reconstruction of the medial patellofemoral ligament (MPFL), and combined procedures. METHODS: In 66 consecutive patients (age 22.8 ± 6.4years) with MACI at the patella (n = 16), MPFL reconstruction (MPFL; n = 31), or combined procedures (n = 19) 3T MRI was performed 2.5 years after surgery. For morphological MRI evaluation WORMS and MOCART scores were obtained. In addition quantitative cartilage T2 and T1rho relaxation times were acquired. Several clinical scores were obtained. Statistical analyses included descriptive statistics, Mann-Whitney-U-tests and Pearson correlations. RESULTS: WORMS scores at follow-up (FU) were significantly worse after combined procedures (8.7 ± 4.9) than after isolated MACI (4.3 ± 3.6, P = 0.005) and after isolated MPFL reconstruction (5.3 ± 5.7, P = 0.004). Bone marrow edema at the patella in the combined group was the only (non-significantly) worsening WORMS parameter from pre- to postoperatively. MOCART scores were significantly worse in the combined group than in the isolated MACI group (57 ± 3 vs 88 ± 9, P < 0.001). Perfect defect filling was achieved in 26% and 69% of cases in the combined and MACI group, respectively (P = 0.031). Global and patellar T2 values were higher in the combined group (Global T2: 34.0 ± 2.8ms) and MACI group (35.5 ± 3.1ms) as compared to the MPFL group (31.1 ± 3.2ms, P < 0.05). T2 values correlated significantly with clinical scores (P < 0.005). Clinical Cincinnati scores were significantly worse in the combined group (P < 0.05). CONCLUSION: After combined surgery with patellar MACI and MPFL reconstruction inferior MRI outcomes were observed than after isolated procedures. Therefore, patients with need for combined surgery may be at particular risk for osteoarthritis.

Evaluation of a deep learning-based reconstruction method for denoising and image enhancement of shoulder MRI in patients with shoulder pain.

OBJECTIVES: To evaluate the diagnostic performance of an automated reconstruction algorithm combining MR imaging acquired using compressed SENSE (CS) with deep learning (DL) in order to reconstruct denoised high-quality images from undersampled MR images in patients with shoulder pain. METHODS: Prospectively, thirty-eight patients (14 women, mean age 40.0 ± 15.2 years) with shoulder pain underwent morphological MRI using a pseudo-random, density-weighted k-space scheme with an acceleration factor of 2.5 using CS only. An automated DL-based algorithm (CS DL) was used to create reconstructions of the same k-space data as used for CS reconstructions. Images were analyzed by two radiologists and assessed for pathologies, image quality, and visibility of anatomical landmarks using a 4-point Likert scale. RESULTS: Overall agreement for the detection of pathologies between the CS DL reconstructions and CS images was substantial to almost perfect (κ 0.95 (95% confidence interval 0.82-1.00)). Image quality and the visibility of the rotator cuff, articular cartilage, and axillary recess were overall rated significantly higher for CS DL images compared to CS (p < 0.03). Contrast-to-noise ratios were significantly higher for cartilage/fluid (CS DL 198 ± 24.3, CS 130 ± 32.2, p = 0.02) and ligament/fluid (CS DL 184 ± 17.3, CS 141 ± 23.5, p = 0.03) and SNR values were significantly higher for ligaments and muscle of the CS DL reconstructions (p < 0.04). CONCLUSION: Evaluation of shoulder pathologies was feasible using a DL-based algorithm for MRI reconstruction and denoising. In clinical routine, CS DL may be beneficial in particular for reducing image noise and may be useful for the detection and better discrimination of discrete pathologies. Assessment of shoulder pathologies was feasible with improved image quality as well as higher SNR using a compressed sensing deep learning-based framework for image reconstructions and denoising. KEY POINTS: • Automated deep learning-based reconstructions showed a significant increase in signal-to-noise ratio and contrast-to-noise ratio (p < 0.04) with only a slight increase of reconstruction time of 40 s compared to CS. • All pathologies were accurately detected with no loss of diagnostic information or prolongation of the scan time. • Significant improvements of the image quality as well as the visibility of the rotator cuff, articular cartilage, and axillary recess were detected.

Assessment of glenoid bone loss and other osseous shoulder pathologies comparing MR-based CT-like images with conventional CT.

OBJECTIVES: To evaluate and compare the diagnostic performance of CT-like images based on a 3D T1-weighted spoiled gradient-echo sequence (T1 GRE), an ultra-short echo time sequence (UTE), and a 3D T1-weighted spoiled multi-echo gradient-echo sequence (FRACTURE) with conventional CT in patients with suspected osseous shoulder pathologies. MATERIALS AND METHODS: Patients with suspected traumatic dislocation of the shoulder (n = 46, mean age 40 ± 14.5 years, 19 women) were prospectively recruited and received 3-T MR imaging including 3D T1 GRE, UTE, and 3D FRACTURE sequences. CT was performed in patients with acute fractures and served as standard of reference (n = 25). Agreement of morphological features between the modalities was analyzed including the glenoid bone loss, Hill-Sachs interval, glenoid track, and the anterior straight-line length. Agreement between the modalities was assessed using Bland-Altman plots, Student's t-test, and Pearson's correlation coefficient. Inter- and intrareader assessment was evaluated with weighted Cohen's κ and intraclass correlation coefficient. RESULTS: All osseous pathologies were detected accurately on all three CT-like sequences (n = 25, κ = 1.00). No significant difference in the percentage of glenoid bone loss was found between CT (mean ± standard deviation, 20.3% ± 8.0) and CT-like MR images (FRACTURE 20.6% ± 7.9, T1 GRE 20.4% ± 7.6, UTE 20.3% ± 7.7, p > 0.05). When comparing the different measurements on CT-like images, measurements performed using the UTE images correlated best with CT. CONCLUSION: Assessment of bony Bankart lesions and other osseous pathologies was feasible and accurate using CT-like images based on 3-T MRI compared with conventional CT. Compared to the T1 GRE and FRACTURE sequence, the UTE measurements correlated best with CT. CLINICAL RELEVANCE STATEMENT: In an acute trauma setting, CT-like images based on a T1 GRE, UTE, or FRACTURE sequence might be a useful alternative to conventional CT scan sparing associated costs as well as radiation exposure. KEY POINTS: • No significant differences were found for the assessment of the glenoid bone loss when comparing measurements of CT-like MR images with measurements of conventional CT images. • Compared to the T1 GRE and FRACTURE sequence, the UTE measurements correlated best with CT whereas the FRACTURE sequence appeared to be the most robust regarding motion artifacts. • The T1 GRE sequence had the highest resolution with high bone contrast and detailed delineation of even small fractures but was more susceptible to motion artifacts.

Diagnostic value of water-fat-separated images and CT-like susceptibility-weighted images extracted from a single ultrashort echo time sequence for the evaluation of vertebral fractures and degenerative changes of the spine.

OBJECTIVES: To evaluate the performance of single-echo Dixon water-fat imaging and computed tomography (CT)-like imaging based on a single ultrashort echo time (sUTE) MR sequence for imaging of vertebral fractures as well as degenerative bone changes of the spine in comparison to conventional CT and MR sequences. METHODS: Thirty patients with suspected acute vertebral fractures were examined using a 3-T MRI, including an sUTE sequence as well as short-tau inversion recovery (STIR) and T1-weighted sequences. During postprocessing, water-fat separation was performed by solving the smoothness-constrained inverse water-fat problem based on a single-complex UTE image. By removing the unwanted low-frequency phase terms, additional MR-based susceptibility-weighted-like (SW-like) images with CT-like contrast were created. Two radiologists evaluated semi-quantitative and quantitative features of fractures and degenerative changes independently and separately on CT and MR images. RESULTS: In total, all 58 fractures were accurately detected of whom 24 were correctly classified as acute fractures with an edema detected on the water-fat-separated UTE images, using STIR and T1w sequences as standard of reference. For the morphological assessment of fractures and degenerative changes, the overall agreement between SW-like images and CT was substantial to excellent (e.g., Genant: κ 0.90 (95% confidence interval 0.54-1.00); AO/Magerl: κ 0.75 (95% confidence interval 0.43-1.00)). Overall inter-reader agreement for water-fat-separated UTE images and SW-like images was substantial to almost perfect. CONCLUSION: Detection and assessment of vertebral fractures and degenerative bone changes of the spine were feasible and accurate using water-fat-separated images as well as SW-like images, both derived from the same sUTE-Dixon sequence. KEY POINTS: • The detection of acute vertebral fractures was feasible using water-fat-separated images and CT-like images reconstructed from one sUTE sequence. • Assessment of the vertebral fractures using SW-like images with CT-like contrast was found to be comparable to conventional CT. • sUTE imaging of the spine can help reduce examination times and radiation exposure.

Differentiation of benign and malignant vertebral fractures using a convolutional neural network to extract CT-based texture features.

PURPOSE: To assess the diagnostic performance of three-dimensional (3D) CT-based texture features (TFs) using a convolutional neural network (CNN)-based framework to differentiate benign (osteoporotic) and malignant vertebral fractures (VFs). METHODS: A total of 409 patients who underwent routine thoracolumbar spine CT at two institutions were included. VFs were categorized as benign or malignant using either biopsy or imaging follow-up of at least three months as standard of reference. Automated detection, labelling, and segmentation of the vertebrae were performed using a CNN-based framework ( https://anduin.bonescreen.de ). Eight TFs were extracted: Varianceglobal, Skewnessglobal, energy, entropy, short-run emphasis (SRE), long-run emphasis (LRE), run-length non-uniformity (RLN), and run percentage (RP). Multivariate regression models adjusted for age and sex were used to compare TFs between benign and malignant VFs. RESULTS: Skewnessglobal showed a significant difference between the two groups when analyzing fractured vertebrae from T1 to L6 (benign fracture group: 0.70 [0.64-0.76]; malignant fracture group: 0.59 [0.56-0.63]; and p = 0.017), suggesting a higher skewness in benign VFs compared to malignant VFs. CONCLUSION: Three-dimensional CT-based global TF skewness assessed using a CNN-based framework showed significant difference between benign and malignant thoracolumbar VFs and may therefore contribute to the clinical diagnostic work-up of patients with VFs.

Assessment of vertebral fractures and edema of the thoracolumbar spine based on water-fat and susceptibility-weighted images derived from a single ultra-short echo time scan.

PURPOSE: To develop a methodology to simultaneously perform single echo Dixon water-fat imaging and susceptibility-weighted imaging (SWI) based on a single echo time (TE) ultra-short echo time (UTE) (sUTE) scan to assess vertebral fractures and degenerative bone changes in the thoracolumbar spine. METHODS: A methodology was developed to solve the smoothness-constrained inverse water-fat problem to separate water and fat while removing unwanted low-frequency phase terms. Additionally, the corrected UTE phase was used for SWI. UTE imaging (TE: 0.14 ms, 3T MRI) was performed in the lumbar spine of nine patients with vertebral fractures and bone marrow edema (BME). All images were reviewed by two radiologists. Water- and fat-separated images were analyzed in comparison with short-tau inversion recovery (STIR) and with respect to BME visibility. The visibility of fracture lines and cortical outlining of the UTE magnitude images were analyzed in comparison with computed tomography. RESULTS: Unwanted phase components, dominated by the B1 phase, were removed from the UTE phase images. The rating of the diagnostic quality of BME visualization showed a high preference for the sUTE-Dixon water- and fat-separated images in comparison with STIR. The UTE magnitude images enabled better visualizing fracture lines compared with STIR and slightly better visibility of cortical outlining. With increasing SWI weighting osseous structures and fatty tissues were enhanced. CONCLUSION: The proposed sUTE-Dixon-SWI methodology allows the removal of unwanted low-frequency phases and enables water-fat separation and SWI processing from a single complex UTE image. The methodology can be used for the simultaneous assessment of vertebral fractures and BME of the thoracolumbar spine.

Deep learning-based acceleration of Compressed Sense MR imaging of the ankle.

OBJECTIVES: To evaluate a compressed sensing artificial intelligence framework (CSAI) to accelerate MRI acquisition of the ankle. METHODS: Thirty patients were scanned at 3T. Axial T2-w, coronal T1-w, and coronal/sagittal intermediate-w scans with fat saturation were acquired using compressed sensing only (12:44 min, CS), CSAI with an acceleration factor of 4.6-5.3 (6:45 min, CSAI2x), and CSAI with an acceleration factor of 6.9-7.7 (4:46 min, CSAI3x). Moreover, a high-resolution axial T2-w scan was obtained using CSAI with a similar scan duration compared to CS. Depiction and presence of abnormalities were graded. Signal-to-noise and contrast-to-noise were calculated. Wilcoxon signed-rank test and Cohen's kappa were used to compare CSAI with CS sequences. RESULTS: The correlation was perfect between CS and CSAI2x (κ = 1.0) and excellent for CS and CSAI3x (κ = 0.86-1.0). No significant differences were found for the depiction of structures between CS and CSAI2x and the same abnormalities were detected in both protocols. For CSAI3x the depiction was graded lower (p ≤ 0.001), though most abnormalities were also detected. For CSAI2x contrast-to-noise fluid/muscle was higher compared to CS (p ≤ 0.05), while no differences were found for other tissues. Signal-to-noise and contrast-to-noise were higher for CSAI3x compared to CS (p ≤ 0.05). The high - resolution axial T2-w sequence specifically improved the depiction of tendons and the tibial nerve (p ≤ 0.005). CONCLUSIONS: Acquisition times can be reduced by 47% using CSAI compared to CS without decreasing diagnostic image quality. Reducing acquisition times by 63% is feasible but should be reserved for specific patients. The depiction of specific structures is improved using a high-resolution axial T2-w CSAI scan. KEY POINTS: • Prospective study showed that CSAI enables reduction in acquisition times by 47% without decreasing diagnostic image quality. • Reducing acquisition times by 63% still produces images with an acceptable diagnostic accuracy but should be reserved for specific patients. • CSAI may be implemented to scan at a higher resolution compared to standard CS images without increasing acquisition times.

Diagnosis of joint invasion in patients with malignant bone tumors: value and reproducibility of direct and indirect signs on MR imaging.

OBJECTIVES: To evaluate the performance and reproducibility of MR imaging features in the diagnosis of joint invasion (JI) by malignant bone tumors. METHODS: MR images of patients with and without JI (n = 24 each), who underwent surgical resection at our institution, were read by three radiologists. Direct (intrasynovial tumor tissue (ITT), intraarticular destruction of cartilage/bone, invasion of capsular/ligamentous insertions) and indirect (tumor size, signal alterations of epiphyseal/transarticular bone (bone marrow replacement/edema-like), synovial contrast enhancement, joint effusion) signs of JI were assessed. Odds ratios, sensitivity, specificity, PPV, NPV, and reproducibilities (Cohen's and Fleiss' κ) were calculated for each feature. Moreover, the diagnostic performance of combinations of direct features was assessed. RESULTS: Forty-eight patients (28.7 ± 21.4 years, 26 men) were evaluated. All readers reliably assessed the presence of JI (sensitivity = 92-100 %; specificity = 88-100%, respectively). Best predictors for JI were direct visualization of ITT (OR = 186-229, p < 0.001) and destruction of intraarticular bone (69-324, p < 0.001). Direct visualization of ITT was also highly reliable in assessing JI (sensitivity, specificity, PPV, NPV = 92-100 %), with excellent reproducibility (κ = 0.83). Epiphyseal bone marrow replacement and synovial contrast enhancement were the most sensitive indirect signs, but lacked specificity (29-54%). By combining direct signs with high specificity, sensitivity was increased (96 %) and specificity (100 %) was maintained. CONCLUSION: JI by malignant bone tumors can reliably be assessed on preoperative MR images with high sensitivity, specificity, and reproducibility. Particularly direct visualization of ITT, destruction of intraarticular bone, and a combination of highly specific direct signs were valuable, while indirect signs were less predictive and specific. KEY POINTS: • Direct visualization of intrasynovial tumor was the single most sensitive and specific (92-100%) MR imaging sign of joint invasion. • Indirect signs of joint invasion, such as joint effusion or synovial enhancement, were less sensitive and specific compared to direct signs. • A combination of the most specific direct signs of joint invasion showed best results with perfect specificity and PPV (both 100%) and excellent sensitivity and NPV (both 96 %).

Development and evaluation of machine learning models based on X-ray radiomics for the classification and differentiation of malignant and benign bone tumors.

OBJECTIVES: To develop and validate machine learning models to distinguish between benign and malignant bone lesions and compare the performance to radiologists. METHODS: In 880 patients (age 33.1 ± 19.4 years, 395 women) diagnosed with malignant (n = 213, 24.2%) or benign (n = 667, 75.8%) primary bone tumors, preoperative radiographs were obtained, and the diagnosis was established using histopathology. Data was split 70%/15%/15% for training, validation, and internal testing. Additionally, 96 patients from another institution were obtained for external testing. Machine learning models were developed and validated using radiomic features and demographic information. The performance of each model was evaluated on the test sets for accuracy, area under the curve (AUC) from receiver operating characteristics, sensitivity, and specificity. For comparison, the external test set was evaluated by two radiology residents and two radiologists who specialized in musculoskeletal tumor imaging. RESULTS: The best machine learning model was based on an artificial neural network (ANN) combining both radiomic and demographic information achieving 80% and 75% accuracy at 75% and 90% sensitivity with 0.79 and 0.90 AUC on the internal and external test set, respectively. In comparison, the radiology residents achieved 71% and 65% accuracy at 61% and 35% sensitivity while the radiologists specialized in musculoskeletal tumor imaging achieved an 84% and 83% accuracy at 90% and 81% sensitivity, respectively. CONCLUSIONS: An ANN combining radiomic features and demographic information showed the best performance in distinguishing between benign and malignant bone lesions. The model showed lower accuracy compared to specialized radiologists, while accuracy was higher or similar compared to residents. KEY POINTS: • The developed machine learning model could differentiate benign from malignant bone tumors using radiography with an AUC of 0.90 on the external test set. • Machine learning models that used radiomic features or demographic information alone performed worse than those that used both radiomic features and demographic information as input, highlighting the importance of building comprehensive machine learning models. • An artificial neural network that combined both radiomic and demographic information achieved the best performance and its performance was compared to radiology readers on an external test set.

Evaluation of MR-derived simulated CT-like images and simulated radiographs compared to conventional radiography in patients with shoulder pain: a proof-of-concept study.

BACKGROUND: To evaluate the diagnostic value of MR-derived CT-like images and simulated radiographs compared with conventional radiographs in patients with suspected shoulder pathology. METHODS: 3 T MRI of the shoulder including a 3D T1-weighted gradient echo sequence was performed in 25 patients (mean age 52.4 ± 18 years, 13 women) with suspected shoulder pathology. Subsequently a cone-beam forward projection algorithm was used to obtain intensity-inverted CT-like images and simulated radiographs. Two radiologists evaluated the simulated images separately and independently using the conventional radiographs as the standard of reference, including measurements of the image quality, acromiohumeral distance, critical shoulder angle, degenerative joint changes and the acromial type. Additionally, the CT-like MR images were evaluated for glenoid defects, subcortical cysts and calcifications. Agreement between the MR-derived simulated radiographs and conventional radiographs was calculated using Cohen's Kappa. RESULTS: Measurements on simulated radiographs and conventional radiographs overall showed a substantial to almost perfect inter- and intra-rater agreement (κ = 0.69-1.00 and κ = 0.65-0.85, respectively). Image quality of the simulated radiographs was rated good to excellent (1.6 ± 0.7 and 1.8 ± 0.6, respectively) by the radiologists. A substantial agreement was found regarding diagnostically relevant features, assessed on Y- and anteroposterior projections (κ = 0.84 and κ = 0.69 for the measurement of the CSA; κ = 0.95 and κ = 0.60 for the measurement of the AHD; κ = 0.77 and κ = 0.77 for grading of the Samilson-Prieto classification; κ = 0.83 and κ = 0.67 for the grading of the Bigliani classification, respectively). CONCLUSION: In this proof-of-concept study, clinically relevant features of the shoulder joint were assessed reliably using MR-derived CT-like images and simulated radiographs with an image quality equivalent to conventional radiographs. MR-derived CT-like images and simulated radiographs may provide useful diagnostic information while reducing the amount of radiation exposure.

Patellar instability MRI measurements are associated with knee joint degeneration after reconstruction of the medial patellofemoral ligament.

OBJECTIVE: To qualitatively and quantitatively evaluate the 2-year magnetic resonance imaging (MRI) outcome after MPFL reconstruction at the knee and to assess MRI-based risk factors that predispose for inferior clinical and imaging outcomes. MATERIALS AND METHODS: A total of 31 patients with MPFL reconstruction were included (22 ± 6 years, 10 female). MRI was performed preoperatively in 21/31 patients. Two-year follow-up MRI included quantitative cartilage T2 and T1rho relaxation time measurements at the ipsilateral and contralateral knee. T2relative was calculated as T2patellofemoral/T2femorotibial. Morphological evaluation was conducted via WORMS scores. Patellar instability parameters and clinical scores were obtained. Statistical analyses included descriptive statistics, t-tests, multivariate regression models, and correlation analyses. RESULTS: Two years after MPFL reconstruction, all patellae were clinically stable. Mean total WORMS scores improved significantly from baseline to follow-up (mean difference ± SEM, - 4.0 ± 1.3; P = 0.005). As compared to patients with no worsening of WORMS subscores over time (n = 5), patients with worsening of any WORMS subscore (n = 16) had lower trochlear depth, lower facetal ratio, higher tibial-tuberosity to trochlear groove (TTTG) distance, and higher postoperative lateral patellar tilt (P < 0.05). T2relative was higher at the ipsilateral knee (P = 0.010). T2relative was associated with preoperatively higher patellar tilt (P = 0.021) and higher TTTG distance (P = 0.034). TTTG distance, global T2 values, and WORMS progression correlated with clinical outcomes (P < 0.05). CONCLUSION: MPFL reconstruction is an optimal treatment strategy to restore patellar stability. Still, progressive knee joint degeneration and patellofemoral cartilage matrix degeneration may be observed, with patellar instability MRI parameters representing particular risk factors.

Multitask Deep Learning for Segmentation and Classification of Primary Bone Tumors on Radiographs.

Background An artificial intelligence model that assesses primary bone tumors on radiographs may assist in the diagnostic workflow. Purpose To develop a multitask deep learning (DL) model for simultaneous bounding box placement, segmentation, and classification of primary bone tumors on radiographs. Materials and Methods This retrospective study analyzed bone tumors on radiographs acquired prior to treatment and obtained from patient data from January 2000 to June 2020. Benign or malignant bone tumors were diagnosed in all patients by using the histopathologic findings as the reference standard. By using split-sample validation, 70% of the patients were assigned to the training set, 15% were assigned to the validation set, and 15% were assigned to the test set. The final performance was evaluated on an external test set by using geographic validation, with accuracy, sensitivity, specificity, and 95% CIs being used for classification, the intersection over union (IoU) being used for bounding box placements, and the Dice score being used for segmentations. Results Radiographs from 934 patients (mean age, 33 years ± 19 [standard deviation]; 419 women) were evaluated in the internal data set, which included 667 benign bone tumors and 267 malignant bone tumors. Six hundred fifty-four patients were in the training set, 140 were in the validation set, and 140 were in the test set. One hundred eleven patients were in the external test set. The multitask DL model achieved 80.2% (89 of 111; 95% CI: 72.8, 87.6) accuracy, 62.9% (22 of 35; 95% CI: 47, 79) sensitivity, and 88.2% (67 of 76; CI: 81, 96) specificity in the classification of bone tumors as malignant or benign. The model achieved an IoU of 0.52 ± 0.34 for bounding box placements and a mean Dice score of 0.60 ± 0.37 for segmentations. The model accuracy was higher than that of two radiologic residents (71.2% and 64.9%; P = .002 and P < .001, respectively) and was comparable with that of two musculoskeletal fellowship-trained radiologists (83.8% and 82.9%; P = .13 and P = .25, respectively) in classifying a tumor as malignant or benign. Conclusion The developed multitask deep learning model allowed for accurate and simultaneous bounding box placement, segmentation, and classification of primary bone tumors on radiographs. © RSNA, 2021 Online supplemental material is available for this article. See also the editorial by Carrino in this issue.

Trajectory correction based on the gradient impulse response function improves high-resolution UTE imaging of the musculoskeletal system.

PURPOSE: UTE sequences typically acquire data during the ramping up of the gradient fields, which makes UTE imaging prone to eddy current and system delay effects. The purpose of this work was to use a simple gradient impulse response function (GIRF) measurement to estimate the real readout gradient waveform and to demonstrate that precise knowledge of the gradient waveform is important in the context of high-resolution UTE musculoskeletal imaging. METHODS: The GIRF was measured using the standard hardware of a 3 Tesla scanner and applied on 3D radial UTE data (TE: 0.14 ms). Experiments were performed on a phantom, in vivo on a healthy knee, and in vivo on patients with spine fractures. UTE images were reconstructed twice, first using the GIRF-corrected gradient waveforms and second using nominal-corrected waveforms, correcting for the low-pass filter characteristic of the gradient chain. RESULTS: Images reconstructed with the nominal-corrected gradient waveforms exhibited blurring and showed edge artifacts. The blurring and the edge artifacts were reduced when the GIRF-corrected gradient waveforms were used, as shown in single-UTE phantom scans and in vivo dual-UTE gradient-echo scans in the knee. Further, the importance of the GIRF-based correction was indicated in UTE images of the lumbar spine, where thin bone structures disappeared when the nominal correction was employed. CONCLUSION: The presented GIRF-based trajectory correction method using standard scanner hardware can improve the quality of high-resolution UTE musculoskeletal imaging.

Soft tissue masses: distribution of entities and rate of malignancy in small lesions.

BACKGROUND: Small soft tissue masses are often falsely assumed to be benign and resected with failure to achieve tumor-free margins. Therefore, this study retrospectively investigated the distribution of histopathologic diagnosis to be encountered in small soft tissue tumors (≤ 5 cm) in a large series of a tertiary referral center. METHODS: Patients with a soft tissue mass (STM) with a maximum diameter of 5 cm presenting at our institution over a period of 10 years, who had undergone preoperative Magnetic resonance imaging and consequent biopsy or/and surgical resection, were included in this study. A final histopathological diagnosis was available in all cases. The maximum tumor diameter was determined on MR images by one radiologist. Moreover, tumor localization (head/neck, trunk, upper extremity, lower extremity, hand, foot) and depth (superficial / deep to fascia) were assessed. RESULTS: In total, histopathologic results and MR images of 1753 patients were reviewed. Eight hundred seventy patients (49.63%) showed a STM ≤ 5 cm and were therefore included in this study (46.79 +/- 18.08 years, 464 women). Mean maximum diameter of the assessed STMs was 2.88 cm. Of 870 analyzed lesions ≤ 5 cm, 170 (19.54%) were classified as superficial and 700 (80.46%) as deep. The malignancy rate of all lesions ≤ 5 cm was at 22.41% (superficial: 23.53% / deep: 22.14%). The malignancy rate dropped to 16.49% (20.79% / 15.32%) when assessing lesions ≤ 3 cm (p = 0.007) and to 15.0% (18.18% / 13.79%) when assessing lesions ≤ 2 cm (p = 0.006). Overall, lipoma was the most common benign lesion of superficial STMs (29.41%) and tenosynovial giant cell tumor was the most common benign lesion of deep STMs (23.29%). Undifferentiated pleomorphic sarcoma was the most common malignant diagnosis among both, superficial (5.29%) and deep (3.57%) STMs. CONCLUSIONS: The rate of malignancy decreased significantly with tumor size in both, superficial and deep STMs. The distribution of entities was different between superficial and deep STMs, yet there was no significant difference found in the malignancy rate.

CT-like images based on T1 spoiled gradient-echo and ultra-short echo time MRI sequences for the assessment of vertebral fractures and degenerative bone changes of the spine.

OBJECTIVES: To evaluate the performance of 3D T1w spoiled gradient-echo (T1SGRE) and ultra-short echo time (UTE) MRI sequences for the detection and assessment of vertebral fractures and degenerative bone changes compared with conventional CT. METHODS: Fractures (n = 44) and degenerative changes (n = 60 spinal segments) were evaluated in 30 patients (65 ± 14 years, 18 women) on CT and 3-T MRI, including CT-like images derived from T1SGRE and UTE. Two radiologists evaluated morphological features on both modalities: Genant and AO/Magerl classifications, anterior/posterior vertebral height, fracture age; disc height, neuroforaminal diameter, grades of spondylolisthesis, osteophytes, sclerosis, and facet joint degeneration. Diagnostic accuracy and agreement between MRI and CT and between radiologists were assessed using crosstabs, weighted κ, and intraclass correlation coefficients. Image quality was graded on a Likert scale. RESULTS: For fracture detection, sensitivity, specificity, and accuracy were 0.95, 0.98, and 0.97 for T1SGRE and 0.91, 0.96, and 0.95 for UTE. Agreement between T1SGRE and CT was substantial to excellent (e.g., Genant: κ, 0.92 [95% confidence interval, 0.83-1.00]; AO/Magerl: κ, 0.90 [0.76-1.00]; osteophytes: κ, 0.91 [0.82-1.00]; sclerosis: κ, 0.68 [0.48-0.88]; spondylolisthesis: ICCs, 0.99 [0.99-1.00]). Agreement between UTE and CT was lower, ranging from moderate (e.g., sclerosis: κ, 0.43 [0.26-0.60]) to excellent (spondylolisthesis: ICC, 0.99 [0.99-1.00]). Inter-reader agreement was substantial to excellent (0.52-1.00), respectively, for all parameters. Median image quality of T1SGRE was rated significantly higher than that of UTE (p < 0.001). CONCLUSIONS: Morphologic assessment of bone pathologies of the spine using MRI was feasible and comparable to CT, with T1SGRE being more robust than UTE. KEY POINTS: • Vertebral fractures and degenerative bone changes can be assessed on CT-like MR images, with 3D T1w spoiled gradient-echo-based images showing a high diagnostic accuracy and agreement with CT. • This could enable MRI to precisely assess bone morphology, and 3D T1SGRE MRI sequences may substitute additional spinal CT examinations in the future. • Image quality and robustness of T1SGRE sequences are higher than those of UTE MRI for the assessment of bone structures.

Opportunistic osteoporosis screening: contrast-enhanced dual-layer spectral CT provides accurate measurements of vertebral bone mineral density.

OBJECTIVES: Osteoporosis remains under-diagnosed, which may be improved by opportunistic bone mineral density (BMD) measurements on CT. However, correcting for the influence of intravenous iodine-based contrast agent is challenging. The purpose of this study was to assess the diagnostic accuracy of iodine-corrected vertebral BMD measurements derived from non-dedicated contrast-enhanced phantomless dual-layer spectral CT (DLCT) examinations. METHODS: Vertebral volumetric DLCT-BMD was measured in native, arterial, and portal-venous scans of 132 patients (63 ± 16 years; 32% women) using virtual monoenergetic images (50 and 200 keV). For comparison, conventional BMD was determined using an asynchronous QCT calibration. Additionally, iodine densities were measured in the abdominal aorta (AA), inferior vena cava, and vena portae (VP) on each CT phase to adjust for iodine-related measurement errors in multivariable linear regressions and a generalized estimated equation, and conversion equations were calculated. RESULTS: BMD values derived from contrast-enhanced phases using conversion equations adjusted for individual vessel iodine concentrations of VP and/or AA showed a high agreement with those from non-enhanced scans in Bland-Altman plots. Mean absolute errors (MAE) of DLCT-BMD were 3.57 mg/ml for the arterial (R2 = 0.989) and 3.69 mg/ml for the portal-venous phase (R2 = 0.987) (conventional BMD: 4.70 [R2 = 0.983] and 5.15 mg/ml [R2 = 0.981]). In the phase-independent analysis, MAE was 4.49 mg/ml for DLCT (R2 = 0.989) (conventional BMD: 4.82 mg/ml [R2 = 0.981]). CONCLUSIONS: Converted BMD derived from contrast-enhanced DLCT examinations and adjusted for individual vessel iodine concentrations showed a high agreement with non-enhanced DLCT-BMD, suggesting that opportunistic BMD measurements are feasible even in non-dedicated contrast-enhanced DLCT examinations. KEY POINTS: • Accurate BMD values can be converted from contrast-enhanced DLCT scans, independent from the used scan phase. • DLCT-BMD measurements from contrast-enhanced scans should be adjusted with iodine concentrations of portal vein and/or abdominal aorta, which significantly improves the goodness-of-fit of conversion models.

Meniscal ramp lesions: frequency, natural history, and the effect on knee cartilage over 2 years in subjects with anterior cruciate ligament tears.

PURPOSE: (i) To investigate the frequency and natural evolution of meniscal ramp lesions (MRLs) on MRI in subjects with acute ACL tear and (ii) to compare knee cartilage compositional degeneration between subjects with MRLs and subjects without meniscal pathology over 2 years. MATERIALS AND METHODS: Fifty-seven subjects with ACL tears (32 females; age 32.6 ± 8.3 years; BMI 24.5 ± 3.5 kg/m2) from a prospective study were screened for the presence of MRLs. Morphological (high-resolution 3D fast spin-echo) and compositional (T1ρ and T2 mapping) MRI was performed prior to and 2 years after ACL reconstruction. Follow-up MR images were assessed for changes in the signal intensity of the MRLs and the presence of meniscal tears. Differences of compositional parameters were compared between subjects with MRLs and without meniscal lesions using independent samples t tests. RESULTS: MRLs were found in 16% (9/56) of the subjects with ACL tears at baseline. Only one subject with MRLs developed a posterior horn meniscal tear over 2 years. In 12 knees, no meniscal tears were found, which were defined as controls. Most interestingly, cartilage ∆T1ρ of the medial femur and medial tibia increased significantly more in subjects with MRLs compared with controls (mean difference, MF = 6.0 ± 0.8 vs. 2.3 ± 0.6, p = 0.004, and MT = 4.4 ± 1.4 vs. 0.4 ± 0.6, p = 0.027) and medial femur ∆T2 over 2 years increased significantly more in MRL than in control knees (5.1 ± 2.5 ms vs. 2.2 ± 1.9 ms, p = 0.012). CONCLUSION: Subjects with ACL tear presented MRLs in 16% of cases. Compared with controls without meniscal lesions, knees with MRLs demonstrated accelerated degeneration of cartilage composition in the medial knee compartment over 2 years.