Can AI distinguish a bone radiograph from photos of flowers or cars? Evaluation of bone age deep learning model on inappropriate data inputs
OBJECTIVE:To evaluate the behavior of a publicly available deep convolutional neural network (DCNN) bone age algorithm when presented with inappropriate data inputs in both radiological and non-radiological domains. METHODS:We evaluated a publicly available DCNN-based bone age application. The DCNN was trained on 12,612 pediatric hand radiographs and won the 2017 RSNA Pediatric Bone Age Challenge (concordance of 0.991 with radiologist ground-truth). We used the application to analyze 50 left-hand radiographs (appropriate data inputs) and seven classes of inappropriate data inputs in radiological (i.e., chest radiographs) and non-radiological (i.e., image of street numbers) domains. For each image, we noted if (1) the application distinguished between appropriate and inappropriate data inputs and (2) inference time per image. Mean inference times were compared using ANOVA. RESULTS:The 16Bit Bone Age application calculated bone age for all pediatric hand radiographs with mean inference time of 1.1Â s. The application did not distinguish between pediatric hand radiographs and inappropriate image types, including radiological and non-radiological domains. The application inappropriately calculated bone age for all inappropriate image types, with mean inference time of 1.1Â s for all categories (pâ€‰=â€‰1). CONCLUSION/CONCLUSIONS:A publicly available DCNN-based bone age application failed to distinguish between appropriate and inappropriate data inputs and calculated bone age for inappropriate images. The awareness of inappropriate outputs based on inappropriate DCNN input is important if tasks such as bone age determination are automated, emphasizing the need for appropriate oversight at the data input and verification stage to avoid unrecognized erroneous results.
Artificial intelligence in musculoskeletal imaging: a perspective on value propositions, clinical use, and obstacles
Artificial intelligence and deep learning (DL) offer musculoskeletal radiology exciting possibilities in multiple areas, including image reconstruction and transformation, tissue segmentation, workflow support, and disease detection. Novel DL-based image reconstruction algorithms correcting aliasing artifacts, signal loss, and noise amplification with previously unobtainable effectiveness are prime examples of how DL algorithms deliver promised value propositions in musculoskeletal radiology. The speed of DL-based tissue segmentation promises great efficiency gains that may permit the inclusion of tissue compositional-based information routinely into radiology reports. Similarly, DL algorithms give rise to a myriad of opportunities for workflow improvements, including intelligent and adaptive hanging protocols, speech recognition, report generation, scheduling, precertification, and billing. The value propositions of disease-detecting DL algorithms include reduced error rates and increased productivity. However, more studies using authentic clinical workflow settings are necessary to fully understand the value of DL algorithms for disease detection in clinical practice. Successful workflow integration and management of multiple algorithms are critical for translating the value propositions of DL algorithms into clinical practice but represent a major roadblock for which solutions are critically needed. While there is no consensus about the most sustainable business model, radiology departments will need to carefully weigh the benefits and disadvantages of each commercially available DL algorithm. Although more studies are needed to understand the value and impact of DL algorithms on clinical practice, DL technology will likely play an important role in the future of musculoskeletal imaging.
Artificial intelligence for MRI diagnosis of joints: a scoping review of the current state-of-the-art of deep learning-based approaches
Deep learning-based MRI diagnosis of internal joint derangement is an emerging field of artificial intelligence, which offers many exciting possibilities for musculoskeletal radiology. A variety of investigational deep learning algorithms have been developed to detect anterior cruciate ligament tears, meniscus tears, and rotator cuff disorders. Additional deep learning-based MRI algorithms have been investigated to detect Achilles tendon tears, recurrence prediction of musculoskeletal neoplasms, and complex segmentation of nerves, bones, and muscles. Proof-of-concept studies suggest that deep learning algorithms may achieve similar diagnostic performances when compared to human readers in meta-analyses; however, musculoskeletal radiologists outperformed most deep learning algorithms in studies including a direct comparison. Earlier investigations and developments of deep learning algorithms focused on the binary classification of the presence or absence of an abnormality, whereas more advanced deep learning algorithms start to include features for characterization and severity grading. While many studies have focused on comparing deep learning algorithms against human readers, there is a paucity of data on the performance differences of radiologists interpreting musculoskeletal MRI studies without and with artificial intelligence support. Similarly, studies demonstrating the generalizability and clinical applicability of deep learning algorithms using realistic clinical settings with workflow-integrated deep learning algorithms are sparse. Contingent upon future studies showing the clinical utility of deep learning algorithms, artificial intelligence may eventually translate into clinical practice to assist detection and characterization of various conditions on musculoskeletal MRI exams.
AI MSK clinical applications: orthopedic implants
Artificial intelligence (AI) and deep learning have multiple potential uses in aiding the musculoskeletal radiologist in the radiological evaluation of orthopedic implants. These include identification of implants, characterization of implants according to anatomic type, identification of specific implant models, and evaluation of implants for positioning and complications. In addition, natural language processing (NLP) can aid in the acquisition of clinical information from the medical record that can help with tasks like prepopulating radiology reports. Several proof-of-concept works have been published in the literature describing the application of deep learning toward these various tasks, with performance comparable to that of expert musculoskeletal radiologists. Although much work remains to bring these proof-of-concept algorithms into clinical deployment, AI has tremendous potential toward automating these tasks, thereby augmenting the musculoskeletal radiologist.
Pilot study for treatment of symptomatic shoulder arthritis utilizing cooled radiofrequency ablation: a novel technique
OBJECTIVE:To introduce cooled radiofrequency nerve ablation (C-RFA) as an alternative to managing symptomatically moderate to severe glenohumeral osteoarthritis (OA) in patients who have failed other conservative treatments and who are not surgical candidates or refuse surgery. MATERIAL AND METHODS/METHODS:This prospective pilot study includes a total of 12 patients experiencing chronic shoulder pain from moderate to severe glenohumeral OA. Patients underwent anesthetic blocks of the axillary, lateral pectoral, and suprascapular nerves to determine candidacy for C-RFA treatment. Adequate response after anesthetic block was over 50% immediate pain relief. Once patients were deemed candidates, they underwent C-RFA of the three nerves 2-3Â weeks later. Treatment response was evaluated using the clinically validated American Shoulder and Elbow Surgeons (ASES) score and visual analog scale (VAS) to assess pain, stiffness, and functional activities of daily living. Follow-up outcome scores were collected up to 6Â months after C-RFA procedure. RESULTS:Twelve patients underwent C-RFA procedure for shoulder OA. VAS scores significantly improved from 8.8â€‰Â±â€‰0.6 to 2.2â€‰Â±â€‰0.4 6Â months after the C-RFA treatment (pâ€‰<â€‰0.001). Patient's ASES score results significantly improved in total ASES from 17.2â€‰Â±â€‰6.6 to 65.7â€‰Â±â€‰5.9 (pâ€‰<â€‰0.0005). No major complications arose. No patients received re-treatment or underwent shoulder arthroplasty. CONCLUSION/CONCLUSIONS:Image-guided axillary, lateral pectoral, and suprascapular nerve C-RFA has minimal complications and is a promising alternative to treat chronic shoulder pain and stiffness from glenohumeral arthritis.
Three-dimensional analysis for quantification of knee joint space width with weight-bearing CT: comparison with non-weight-bearing CT and weight-bearing radiography
OBJECTIVE:To compare computer-based 3D-analysis for quantification of the femorotibial joint space width (JSW) using weight-bearing cone beam CT (WB-CT), non-weight-bearing multi-detector CT (NWB-CT), and weight-bearing conventional radiographs (WB-XR). DESIGN/METHODS:Twenty-six participants prospectively underwent NWB-CT, WB-CT, and WB-XR of the knee. For WB-CT and NWB-CT, the average and minimal JSW was quantified by 3D-analysis of the minimal distance of any point of the subchondral tibial bone surface and the femur. Associations with mechanical leg axes and osteoarthritis were evaluated. Minimal JSW of WB-CT was further compared to WB-XR. Two-tailed p-values of <0.05 were considered significant. RESULTS:Significant differences existed of the average medial and lateral JSW between WB-CT and NWB-CT (medial: 4.7 vs 5.1Â mm [PÂ =Â 0.028], lateral: 6.3 vs 6.8Â mm [PÂ =Â 0.008]). The minimal JSW on WB-XR (medial:3.1Â mm, lateral:5.8Â mm) were significantly wider compared to WB-CT and NWB-CT (both medial:1.8Â mm, lateral:2.9Â mm, all pÂ <Â 0.001), but not significantly different between WB-CT and NWB-CT (all pÂ â‰¥Â 0.869). Significant differences between WB-CT and NWB-CT existed in participants with varus knee alignment for the average and the minimal medial JSW (pÂ =Â 0.004 and pÂ =Â 0.011) and for participants with valgus alignment for the average lateral JSW (pÂ =Â 0.013). On WB-CT, 25% of the femorotibial compartments showed bone-on-bone apposition, which was significantly higher when compared to NWB-CT (10%,PÂ =Â 0.008) and WB-XR (8%,PÂ =Â 0.012). CONCLUSION/CONCLUSIONS:Combining WB-CT with 3D-based assessment allows detailed quantification of the femorotibial joint space and the effect of knee alignment on JSW. WB-CT demonstrates significantly more bone-on-bone appositions, which are underestimated or even undetectable on NWB-CT and WB-XR.
Interdisciplinary consensus statements on imaging of scapholunate joint instability
OBJECTIVES/OBJECTIVE:The purpose of this agreement was to establish evidence-based consensus statements on imaging of scapholunate joint (SLJ) instability by an expert group using the Delphi technique. METHODS:Nineteen hand surgeons developed a preliminary list of questions on SLJ instability. Radiologists created statements based on the literature and the authors' clinical experience. Questions and statements were revised during three iterative Delphi rounds. Delphi panellists consisted of twenty-seven musculoskeletal radiologists. The panellists scored their degree of agreement to each statement on an eleven-item numeric scale. Scores of '0', '5' and '10' reflected complete disagreement, indeterminate agreement and complete agreement, respectively. Group consensus was defined as a score of '8' or higher for 80% or more of the panellists. RESULTS:Ten of fifteen statements achieved group consensus in the second Delphi round. The remaining five statements achieved group consensus in the third Delphi round. It was agreed that dorsopalmar and lateral radiographs should be acquired as routine imaging work-up in patients with suspected SLJ instability. Radiographic stress views and dynamic fluoroscopy allow accurate diagnosis of dynamic SLJ instability. MR arthrography and CT arthrography are accurate for detecting scapholunate interosseous ligament tears and articular cartilage defects. Ultrasonography and MRI can delineate most extrinsic carpal ligaments, although validated scientific evidence on accurate differentiation between partially or completely torn or incompetent ligaments is not available. CONCLUSIONS:Delphi-based agreements suggest that standardized radiographs, radiographic stress views, dynamic fluoroscopy, MR arthrography and CT arthrography are the most useful and accurate imaging techniques for the work-up of SLJ instability. KEY POINTS/CONCLUSIONS:â€¢ Dorsopalmar and lateral wrist radiographs remain the basic imaging modality for routine imaging work-up in patients with suspected scapholunate joint instability. â€¢ Radiographic stress views and dynamic fluoroscopy of the wrist allow accurate diagnosis of dynamic scapholunate joint instability. â€¢ Wrist MR arthrography and CT arthrography are accurate for determination of scapholunate interosseous ligament tears and cartilage defects.
MRI nomenclature for musculoskeletal infection
The Society of Skeletal Radiology (SSR) Practice Guidelines and Technical Standards Committee identified musculoskeletal infection as a White Paper topic, and selected a Committee, tasked with developing a consensus on nomenclature for MRI of musculoskeletal infection outside the spine. The objective of the White Paper was to critically assess the literature and propose standardized terminology for imaging findings of infection on MRI, in order to improve both communication with clinical colleagues and patient care.A definition was proposed for each term; debate followed, and the committee reached consensus. Potential controversies were raised, with formulated recommendations. The committee arrived at consensus definitions for cellulitis, soft tissue abscess, and necrotizing infection, while discouraging the nonspecific term phlegmon. For bone infection, the term osteitis is not useful; the panel recommends using terms that describe the likelihood of osteomyelitis in cases where definitive signal changes are lacking. The work was presented virtually to SSR members, who had the opportunity for review and modification prior to submission for publication.
[Sclerosing epithelioid fibrosarcoma: A rare pathologic entity]
CT hepatic arterial perfusion index does not allow stratification of the degree of esophageal varices and bleeding risk in cirrhotic patients in Child-Pugh classes A and B
PURPOSE:To evaluate if the hepatic arterial perfusion index (HPI) in liver parenchyma of cirrhotic patients can serve as a surrogate parameter for stratifying the degree of esophageal varices and related bleeding risks. METHODS:CT image data of sixty-six patients (59 men; mean age 68Â yearsâ€‰Â±â€‰10Â years) with liver cirrhosis (Child-Pugh class A (35/66, 53%), B (25/66, 38%), and C (6/66, 9%) who underwent perfusion CT (PCT) for hepatocellular carcinoma (HCC) screening between April 2010 and January 2019 were retrospectively identified. HPI, a parameter calculated by a commercially available CT liver perfusion analysis software that is based on the double maximum slope model, using time attenuation curve to determine perfusion, was correlated with the degree of esophageal varices diagnosed at endoscopy and the number of bleeding events. RESULTS:Eta correlation coefficient for HPI/presence of esophageal varices was very weak (0.083). Spearman-Rho for HPI/grading of esophageal varices was very weak (0.037 (pâ€‰=â€‰0.804)). Kendall-Tau-b for HPI/grading of esophageal varices was very weak (0.027 (pâ€‰=â€‰0.807)). ANOVA and Bonferroni post-hoc-tests showed no significant difference of HPI between different grades of esophageal varices (F (3, 62)â€‰=â€‰1.676, pâ€‰=â€‰0.186). Eta correlation coefficient for HPI/bleeding event was very weak (0.126). CONCLUSION:The stratification of the degree of esophageal varices and the related bleeding risk by correlation with the HPI as a surrogate parameter for portal venous hypertension was not possible for patients with liver cirrhosis in Child-Pugh class A and B.