A deep learning model for the diagnosis of sacroiliitis according to Assessment of SpondyloArthritis International Society classification criteria with magnetic resonance imaging.

PURPOSE: The purpose of this study was to develop and evaluate a deep learning model to detect bone marrow edema (BME) in sacroiliac joints and predict the MRI Assessment of SpondyloArthritis International Society (ASAS) definition of active sacroiliitis in patients with chronic inflammatory back pain. MATERIALS AND METHODS: MRI examinations of patients from the French prospective multicenter DESIR cohort (DEvenir des Spondyloarthropathies Indifférenciées Récentes) were used for training, validation and testing. Patients with inflammatory back pain lasting three months to three years were recruited. Test datasets were from MRI follow-ups at five years and ten years. The model was evaluated using an external test dataset from the ASAS cohort. A neuronal network classifier (mask-RCNN) was trained and evaluated for sacroiliac joints detection and BME classification. Diagnostic capabilities of the model to predict ASAS MRI active sacroiliitis (BME in at least two half-slices) were assessed using Matthews correlation coefficient (MCC), sensitivity, specificity, accuracy and AUC. The gold standard was experts' majority decision. RESULTS: A total of 256 patients with 362 MRI examinations from the DESIR cohort were included, with 27% meeting the ASAS definition for experts. A total of 178 MRI examinations were used for the training set, 25 for the validation set and 159 for the evaluation set. MCCs for DESIR baseline, 5-years, and 10-years follow-up were 0.90 (n = 53), 0.64 (n = 70), and 0.61 (n = 36), respectively. AUCs for predicting ASAS MRI were 0.98 (95% CI: 0.93-1), 0.90 (95% CI: 0.79-1), and 0.80 (95% CI: 0.62-1), respectively. The ASAS external validation cohort included 47 patients (mean age 36 ± 10 [SD] years; women, 51%) with 19% meeting the ASAS definition. MCC was 0.62, sensitivity 56% (95% CI: 42-70), specificity 100% (95% CI: 100-100) and AUC 0.76 (95% CI: 0.57-0.95). CONCLUSION: The deep learning model achieves performance close to those of experts for BME detection in sacroiliac joints and determination of active sacroiliitis according to the ASAS definition.

Assessment of performances of a deep learning algorithm for the detection of limbs and pelvic fractures, dislocations, focal bone lesions, and elbow effusions on trauma X-rays.

PURPOSE: To appraise the performances of an AI trained to detect and localize skeletal lesions and compare them to the routine radiological interpretation. METHODS: We retrospectively collected all radiographic examinations with the associated radiologists' reports performed after a traumatic injury of the limbs and pelvis during 3 consecutive months (January to March 2017) in a private imaging group of 14 centers. Each examination was analyzed by an AI (BoneView, Gleamer) and its results were compared to those of the radiologists' reports. In case of discrepancy, the examination was reviewed by a senior skeletal radiologist to settle on the presence of fractures, dislocations, elbow effusions, and focal bone lesions (FBL). The lesion-wise sensitivity of the AI and the radiologists' reports was compared for each lesion type. This study received IRB approval (CRM-2106-177). RESULTS: A total of 4774 exams were included in the study. Lesion-wise sensitivity was 73.7% for the radiologists' reports vs. 98.1% for the AI (+24.4 points) for fracture detection, 63.3% vs. 89.9% (+26.6 points) for dislocation detection, 84.7% vs. 91.5% (+6.8 points) for elbow effusion detection, and 16.1% vs. 98.1% (+82 points) for FBL detection. The specificity of the radiologists' reports was always 100% whereas AI specificity was 88%, 99.1%, 99.8%, 95.6% for fractures, dislocations, elbow effusions, and FBL respectively. The NPV was measured at 99.5% for fractures, 99.8% for dislocations, and 99.9% for elbow effusions and FBL. CONCLUSION: AI has the potential to prevent diagnosis errors by detecting lesions that were initially missed in the radiologists' reports.

Assessment of an AI Aid in Detection of Adult Appendicular Skeletal Fractures by Emergency Physicians and Radiologists: A Multicenter Cross-sectional Diagnostic Study.

Background The interpretation of radiographs suffers from an ever-increasing workload in emergency and radiology departments, while missed fractures represent up to 80% of diagnostic errors in the emergency department. Purpose To assess the performance of an artificial intelligence (AI) system designed to aid radiologists and emergency physicians in the detection and localization of appendicular skeletal fractures. Materials and Methods The AI system was previously trained on 60 170 radiographs obtained in patients with trauma. The radiographs were randomly split into 70% training, 10% validation, and 20% test sets. Between 2016 and 2018, 600 adult patients in whom multiview radiographs had been obtained after a recent trauma, with or without one or more fractures of shoulder, arm, hand, pelvis, leg, and foot, were retrospectively included from 17 French medical centers. Radiographs with quality precluding human interpretation or containing only obvious fractures were excluded. Six radiologists and six emergency physicians were asked to detect and localize fractures with (n = 300) and fractures without (n = 300) the aid of software highlighting boxes around AI-detected fractures. Aided and unaided sensitivity, specificity, and reading times were compared by means of paired Student t tests after averaging of performances of each reader. Results A total of 600 patients (mean age ± standard deviation, 57 years ± 22; 358 women) were included. The AI aid improved the sensitivity of physicians by 8.7% (95% CI: 3.1, 14.2; P = .003 for superiority) and the specificity by 4.1% (95% CI: 0.5, 7.7; P < .001 for noninferiority) and reduced the average number of false-positive fractures per patient by 41.9% (95% CI: 12.8, 61.3; P = .02) in patients without fractures and the mean reading time by 15.0% (95% CI: -30.4, 3.8; P = .12). Finally, stand-alone performance of a newer release of the AI system was greater than that of all unaided readers, including skeletal expert radiologists, with an area under the receiver operating characteristic curve of 0.94 (95% CI: 0.92, 0.96). Conclusion The artificial intelligence aid provided a gain of sensitivity (8.7% increase) and specificity (4.1% increase) without loss of reading speed. © RSNA, 2021 Online supplemental material is available for this article.

Assessment of typical SpA lesions on MRI of the spine: do local readers and central readers agree in the DESIR-cohort at baseline?

Comparing local reading (LocR) with central reading (CentR) of typical spondyloarhritis lesions including bone marrow edema (BME) and structural lesions on magnetic resonance imaging of the spine (MRI-spine), in patients with inflammatory back pain (IBP; ≥3 months, <3 years). Baseline data of 667 patients, age 18-50 years, from the Devenir des Spondylarthopathies Indifferenciees Recentes (DESIR)-cohort were used. Two trained central readers scored anterior and posterior corner BME, fatty lesions, erosions and syndesmophytes on MRI-spine. Presences of lesions, based on average scores, were used for CentR. A local radiologist and/or rheumatologist scored MRI-spine on presence/doubt/absence of 'inflammation' and 'structural lesions'. Agreement between central readers and readings was calculated (Cohen's kappa: κ). Agreement between central readers was moderate (BME κ = 0.55, fatty lesions κ = 0.50) to slight (erosions κ = 0.12, syndesmophytes κ = 0.19). Agreement between LocR and CentR was κ = 0.32 (BME) and κ = 0.13 (structural lesions). In 78/160 patients (48.8%) LocR were in doubt while CentR scored BME lesions, for structural lesions this was 17.8% (28/157 patients). Agreement between 2 central readers for scoring spondyloarhritis-like lesions on MRI-spine was moderate but better compared to LocR and CentR agreement. LocR often doubt about the presence of MRI-spine lesions while central trained readers score lesions.

Evaluation of a method for the assessment of anterior acetabular coverage and hip joint space width.

OBJECTIVE: The false-profile view (FP) is an oblique view of the acetabulum and a true lateral view of femur and joint. It evaluates anterior and posterior hip joint space width (JSW) and anterior acetabular coverage using the vertical-center margin angle (VCA). The biplanar slot scanner (SS) allows simultaneous bilateral acquisitions of oblique views of hip joints. The aim of this work was to compare SS versus FP for the evaluation of VCA and JSW and to test its reproducibility and validity. MATERIALS AND METHODS: A prospective study of 28 patients (55 hips) with hip pain was performed from November 2011 until May 2012. Two readers measured VCA and JSW. JSW was normalized by the diameter of the femoral head for each technique. The radiation exposure was recorded and compared between the two modalities. Student's t test and the Pearson's correlation assessed the agreement between SS and FP. The intraclass correlation coefficient (ICC) was used to assess the interobserver agreement. RESULTS: The mean VCA angle was 32.1° (± 7.1°) and 30.3° (± 8.5°) with FP and SS, respectively. The coefficient of correlation was 0.90 (p < 0.01). The coefficient of correlation of normalized JSW was 0.83, 0.85, and 0.87 at anterior, vertical, and posterior points, respectively). The ICC was between 0.69 and 0.81 for each modality. The mean radiation exposure was 1.33 (± 0.02) mGy versus 8.69 (±0.04) mGy for FP and SS, respectively (p < 0.0001). CONCLUSIONS: SS has the potential advantages of simultaneous bilateral acquisition, higher standardization, and is less irradiating. SS is reliable for coxometry.

[Morphological imaging for response assessment in metastatic bone disease]. / Imagerie conventionnelle : évaluation de la réponse et impact des traitements.

Response assessment with morphological imaging in metastatic bone disease is difficult. X-Ray is useful only for long bones. CT is the best imaging modality for assessment of osteolysis and osteosclerosis. MRI is the gold standard for bone marrow morphological imaging. Morphological MR imaging using basic sequences, especially T1, provides confident evaluation of the behaviour of lesions under therapy. Radiotherapy and interventional radiology may induce cortical and bone marrow changes.

Assessment of ankylosing spondylitis criteria in patients with chronic low back pain and vertebral endplate Modic I signal changes.

OBJECTIVE: Patients with chronic low back pain (cLBP) and vertebral endplate Modic I signal changes on lumbar magnetic resonance imaging (MRI) have clinical features that could mimic inflammatory back pain related to spondyloarthritis (SpA) and/or ankylosing spondylitis (AS). We aimed to assess whether such patients fulfilled criteria for SpA and/or AS. METHODS: For 5 months in 2008, all patients (n = 314) referred to a tertiary care physical medicine and rehabilitation facility in France were consecutively screened. A total of 185 hospitalized for non-specific cLBP were prospectively assessed. Forty patients fulfilling inclusion criteria were consecutively enrolled and included in 2 groups according to MRI findings: Modic I (n = 15) and non-Modic I (n = 25). MRI findings were assessed independently by 2 spine specialists and a radiologist. HLA-B27 status was determined. Data were collected on clinical measurements and fulfillment of Amor criteria (AC) and modified New York criteria (mNYC). All assessors were blinded to HLA-B27 status. RESULTS: Whatever the Modic group, no patient fulfilled AC or mNYC, and mean total scores were comparable [3 ± 2 (range 0-22; p = 0.977), 1 ± 1 (range 0-3; p = 1.000), and 0 ± 0 (range 0-1; p = 1.000) for AC and clinical and radiological mNYC, respectively]. HLA-B27 status was similar in both groups [n = 2 (13%) vs n = 0 (0%); p = 0.135]. CONCLUSION: Patients with cLBP and Modic I vertebral endplate signal changes on lumbar MRI do not fulfill widely used and validated criteria for SpA and/or AS. Such cases are clinically distinct from SpA and AS.

Assessment of vascular invasion by bone and soft tissue tumours of the limbs: usefulness of MDCT angiography.

OBJECTIVE: To evaluate the accuracy of computed tomography angiography (CTA) in predicting arterial encasement by limb tumours, by comparing CTA with surgical findings (gold standard). METHODS: Preoperative CTA images of 55 arteries in 48 patients were assessed for arterial status: cross-sectional CTA images were scored as showing a fat plane between artery and tumour (score 0), slight contact between artery and tumour (score 1), partial arterial encasement (score 2) or total arterial encasement (score 3). Reformatted CTA images were assessed for arterial displacement, rigid wall, stenosis or occlusion. At surgery, arteries were classified as free or surgically encased; 45 arteries were free and 10 were surgically encased. RESULTS: Multivariate logistic regression identified the axial CTA score as a relevant predictor for arterial encasement and subsequent vascular intervention during surgery. All sites where CTA showed a fat plane between the tumour and the artery were classified as free at surgery (n = 28/28). The sensitivity of total arterial encasement on CTA (score 3) was 90%, specificity 93%, accuracy 93% and positive likelihood ratio 13.5. CONCLUSION: CTA evidence of total arterial encasement is a highly specific indication of arterial encasement. The presence of fat between the tumour and the artery on CTA rules out arterial involvement at surgery.

MRI assessment of recurrent carpal tunnel syndrome after open surgical release of the median nerve.

OBJECTIVE: The purpose of this study was to retrospectively determine the accuracy of MRI in identification of the morphologic features of median nerve dysfunction after surgical release of the median nerve for carpal tunnel syndrome. MATERIALS AND METHODS: Two blinded readers independently evaluated axial 1.5-T MR images for retinacular regrowth, morphologic characteristics of the median nerve, and presence of mass effect, fibrosis, and carpal tunnel decompression. All 47 patients (11 men, 36 women; mean age, 55 years; range, 27-81 years) had undergone open surgical release of the median nerve for carpal tunnel syndrome. Thirty-five patients had electromyographic evidence of recurrent carpal tunnel syndrome. The other 12 patients did not have electrophysiologic evidence of recurrent carpal tunnel syndrome and were the control group. RESULTS: A statistically significant difference between the recurrent carpal tunnel syndrome and control groups was found for fibrosis (p = 0.009), nerve enhancement (p = 0.04), and median nerve width (p = 0.008) and ratio (p = 0.01) at the pisiform level. CONCLUSION: MRI may be used in association with electromyography for accurate postoperative evaluation of the carpal tunnel.

Assessment of vascular invasion by musculoskeletal tumors of the limbs: use of contrast-enhanced MR angiography.

PURPOSE: To prospectively evaluate the accuracy of contrast material-enhanced magnetic resonance (MR) angiography in the evaluation of vascular invasion by bone and soft-tissue tumors, with surgery serving as the reference standard. MATERIALS AND METHODS: This study was approved by the regional ethics committee, and all patients gave informed consent. Preoperative MR angiograms and MR images of 31 sites in 30 patients with bone or soft-tissue sarcomas (n = 21) or other tumors (n = 9) were assessed for features of vascular invasion. All images were prospectively evaluated by two musculoskeletal radiologists working in consensus. MR images were evaluated for the presence of a fat plane between the vessels and the tumor and partial or total encasement of vessels. MR angiograms were evaluated for the presence of vascular displacement, stenosis, or occlusion. MR imaging and MR angiographic features of vascular invasion were graded. Imaging findings were correlated with surgical findings and classified as negative if there was no vascular invasion and as positive if there was vascular invasion. RESULTS: Among the 31 cases, 20 were classified as negative and 11 were classified as positive at surgery. All but three cases with a gap between the tumor and the vessels on MR images were classified as free and without adhesions at surgery. All cases with arterial stenoses at MR angiography had tumoral adhesion or tumoral encasement at surgery. MR imaging had a sensitivity of 64%, a specificity of 95%, a positive predictive value of 88% a negative predictive value of 83%, and an accuracy of 84% in the detection of vascular invasion on the basis of findings of partial or total encasement. MR angiography had a sensitivity of 82%, a specificity of 85%, a positive predictive value of 75%, a negative predictive value of 90%, and an accuracy of 84% in the detection of vascular invasion on the basis of the findings of a stenosis. CONCLUSION: On contrast-enhanced MR angiograms, findings of stenosis were sensitive and specific in the detection of arterial invasion. MR imaging evidence of partial or total encasement is highly specific in the detection of vascular invasion, while MR imaging evidence of a gap between the tumor and the vessels excludes an arterial invasion.