Richard Kijowski

MR-Transformer, a vision transformer-based deep learning model, exhibited state-of-the-art performance in predicting the progression of knee osteoarthritis to total knee replacement using MRI scans from the Osteoarthritis Initiative and Multicenter Osteoarthritis Study databases.

Background: Large language models (LLMs) have shown promising abilities in text-based clinical tasks but they do not inherently interpret medical images such as knee radiographs. Purpose: To develop a human-artificial intelligence interactive diagnostic approach, named radiology generative pretrained transformer (RadGPT), aimed at assisting and synergizing with human users for the interpretation of knee radiological images. Materials and Methods: A total of 22 512 knee roentgen ray images and reports were retrieved from Massachusetts General Hospital; 80% of these were used for model training and 10% were used for model testing and validation, respectively. Fifteen diagnostic imaging features (eg, osteoarthritis, effusion, joint space narrowing, osteophyte) were selected to label images based on their high frequency and clinical relevance in the retrieved official reports. Area under the curve scores were calculated for each feature to assess the diagnostic performance. To evaluate the quality of the generated medical text, historical clinical reports were used as the reference text. Several metrics for text generation tasks are applied, including BiLingual Evaluation Understudy, Recall-Oriented Understudy for Gisting Evaluation, Metric for Evaluation of Translation with Explicit Ordering, and Semantic Propositional Image Caption Evaluation. Results: RadGPT, in collaboration with human users, achieved area under the curve scores ranging from 0.76 for osteonecrosis to 0.91 for arthroplasty across 15 diagnostic categories for knee conditions. Compared with the baseline LLM method, RadGPT achieved higher scores, specifically 0.18 in BiLingual Evaluation Understudy score, 0.30 in Recall-Oriented Understudy for Gisting Evaluation-L, 0.10 in Metric for Evaluation of Translation with Explicit Ordering, and 0.15 in Semantic Propositional Image Caption Evaluation, which is significantly higher than the baseline LLM method, demonstrating good linguistic overlap and clinical consistency with the reference reports. Conclusion: RadGPT has achieved advanced results in knee roentgen ray image feature recognition, illustrating the potential of LLMs in medical image interpretation. The study establishes a training protocol for developing artificial intelligence-assisted tools specifically focusing on the diagnosis and interpretation of knee radiological images.

Deep learning superresolution simultaneous multislice parallel imaging–accelerated knee MRI provided excellent image quality and improved diagnostic performance for articular cartilage defects while maintaining performance for meniscus and ligament tears.

This article will provide a perspective review of the most extensively investigated deep learning (DL) applications for musculoskeletal disease detection that have the best potential to translate into routine clinical practice over the next decade. Deep learning methods for detecting fractures, estimating pediatric bone age, calculating bone measurements such as lower extremity alignment and Cobb angle, and grading osteoarthritis on radiographs have been shown to have high diagnostic performance with many of these applications now commercially available for use in clinical practice. Many studies have also documented the feasibility of using DL methods for detecting joint pathology and characterizing bone tumors on magnetic resonance imaging (MRI). However, musculoskeletal disease detection on MRI is difficult as it requires multi-task, multi-class detection of complex abnormalities on multiple image slices with different tissue contrasts. The generalizability of DL methods for musculoskeletal disease detection on MRI is also challenging due to fluctuations in image quality caused by the wide variety of scanners and pulse sequences used in routine MRI protocols. The diagnostic performance of current DL methods for musculoskeletal disease detection must be further evaluated in well-designed prospective studies using large image datasets acquired at different institutions with different imaging parameters and imaging hardware before they can be fully implemented in clinical practice. Future studies must also investigate the true clinical benefits of current DL methods and determine whether they could enhance quality, reduce error rates, improve workflow, and decrease radiologist fatigue and burnout with all of this weighed against the costs.

This issue of Seminars in Musculoskeletal Radiology discusses the role of biomarkers in musculoskeletal (MSK) imaging. Biomarkers are defined as quantifiable indicators of biological processes or responses. Imaging biomarkers provide valuable anatomical, pathologic, and functional information using imaging modalities such as radiographs, ultrasonography (US), magnetic resonance imaging (MRI), and computed tomography (CT). Imaging biomarkers can be used to aid disease diagnosis, provide valuable information about disease progression and patient survival, and identify patients who are likely to respond favorably to specific therapies. This issue offers eight comprehensive review articles that describe the role of imaging biomarkers in a wide variety of MSK applications, including the evaluation of cartilage, tumors, bone, body composition, and peripheral nerves. As an introduction to the issue, Ronnie Sebro surveys the role of biomarkers in radiology. Through an exploration of various biomarker types, such as imaging biomarkers, molecular biomarkers, and genetic markers, the review outlines their roles in disease detection, prognosis, prediction, and therapeutic monitoring. The article also discusses the significance of robust study designs, blinding, power and sample size calculations, performance metrics, and statistical methodologies in imaging biomarker research. Frank Roemer and associates discuss the role of imaging biomarkers in osteoarthritis research. The current role of radiography, as well as advances in MRI-based quantitative three-dimensional morphological cartilage assessment and semiquantitative whole-organ joint assessment of osteoarthritis, are reviewed. Although MRI has evolved as the leading imaging method in osteoarthritis research, developments in CT imaging are also described. Finally, experiences from the foundations of the National Institutes of Health Biomarker Working Group consortium biomarker qualification study are described. Maximilian Löffler et al survey the technical advances and current applications in the field of MRI-based biomarkers of cartilage composition. Compositional MRI can detect disease-related and treatment-related changes in articular cartilage through assessment of tissue composition and ultrastructure. Various techniques of compositional MRI are reviewed including T2 mapping, sodium imaging, T1rho mapping, ultrashort echo time imaging with T2*, gadolinium-enhanced T1 mapping, chemical exchange-dependent saturation transfer imaging, and diffusion-weighted imaging. The article also discusses standardization efforts by the Quantitative Imaging Biomarkers Alliance that aims to make compositional MRI of cartilage clinically feasible and comparable. Ali Ghasemi and associates discuss MRI biomarkers for evaluating bone and soft tissue tumors. The article describes conventional imaging methods such as T1-weighted, fluid-sensitive, postcontrast, and chemical shift gradient-echo sequences that are useful for delineating the anatomical extent of tumors, defining their character and assessing the region of prior treatment. Various functional MRI methods are also reviewed, such as diffusion-weighted imaging that assesses tumor cellularity, dynamic contrast-enhanced MRI that can offer valuable information about the temporal pattern of tumor enhancement, and MRI spectroscopy that evaluates tumor chemical composition. Daniel Brandenberger and Lawrence White examine the role of radiomics for evaluating MSK tumors. In radiomics, images are converted to higher dimensional image features that are subsequently extracted, or “mined,” for analysis and lesional characterization. Radiomics offers the potential for discovering novel imaging diagnostic and predictive biomarkers using standard-of-care medical imaging. The article details the core concepts of radiomics and the application of radiomics to date in MSK sarcoma research. Specific challenges related to radiomic studies, viewpoints on clinical adoption, and future perspectives in the field are also reviewed. Saeed Jerban et al discuss the role of MRI biomarkers for characterizing the organic, water, fat, and mineral components of bone. The article describes the use of ultrashort echo time sequences for anatomical assessment of bone and for evaluating the total water, bone water, and pore water of bone. Additional ultrashort echo time applications of bone are also reviewed including tricomponent T2* analysis that can distinguish between bound water and pore water signals and fractions, magnetic transfer methods that can assess fat fraction and quantify the organic matrix, and quantitative susceptibility mapping techniques that provide a surrogate measure of bone mineral density. Connie Chang and associates present the relevant clinical and imaging terminology needed to understand the clinical and research applications of body composition. Imaging biomarkers of bone, muscle, and fat tissues obtained with dual-energy X-ray absorptiometry, CT, MRI, and US are described. The article examines the rationale for imaging of body composition, presents the strengths and weaknesses of currently used methods, and reviews the most relevant clinical applications. Secondary analysis of tissue health “opportunistically” on imaging studies obtained for routine clinical care is also discussed. Alireza Eajazi and colleagues discuss the role of MRI and US for evaluating peripheral nerves. The article describes conventional MRI and US methods that can quantify measures of nerve and fascicular size and signal and detect edema-like signal changes in nerves and fatty and fibrous changes in nerves and muscle. Functional biomarkers are also reviewed, such as apparent diffusion coefficient from diffusion-weighted imaging and fractional anisotropy and tractography from diffusion tensor imaging, that are useful for assessing nerve microstructure and integrity and dynamic contrast-enhanced MRI that...

Currently no disease-modifying osteoarthritis drug has been approved for the treatment of osteoarthritis (OA) that can reverse, hold, or slow the progression of structural damage of OA-affected joints. The reasons for failure are manifold and include the heterogeneity of structural disease of the OA joint at trial inclusion, and the sensitivity of biomarkers used to measure a potential treatment effect.This article discusses the role and potential of different imaging biomarkers in OA research. We review the current role of radiography, as well as advances in quantitative three-dimensional morphological cartilage assessment and semiquantitative whole-organ assessment of OA. Although magnetic resonance imaging has evolved as the leading imaging method in OA research, recent developments in computed tomography are also discussed briefly. Finally, we address the experience from the Foundation for the National Institutes of Health Biomarker Consortium biomarker qualification study and the future role of artificial intelligence.

While musculoskeletal imaging volumes are increasing, there is a relative shortage of subspecialized musculoskeletal radiologists to interpret the studies. Will artificial intelligence (AI) be the solution? For AI to be the solution, the wide implementation of AI-supported data acquisition methods in clinical practice requires establishing trusted and reliable results. This implementation will demand close collaboration between core AI researchers and clinical radiologists. Upon successful clinical implementation, a wide variety of AI-based tools can improve the musculoskeletal radiologist's workflow by triaging imaging examinations, helping with image interpretation, and decreasing the reporting time. Additional AI applications may also be helpful for business, education, and research purposes if successfully integrated into the daily practice of musculoskeletal radiology. The question is not whether AI will replace radiologists, but rather how musculoskeletal radiologists can take advantage of AI to enhance their expert capabilities.

Background: Three‐dimensional MR fingerprinting (3D‐MRF) techniques have been recently described for simultaneous multiparametric mapping of knee cartilage. However, investigation of repeatability remains limited.Purpose: To assess the intra‐day and inter‐day repeatabilities of knee cartilage T₁, T₂, and T_1ρ maps using a 3D‐MRF sequence for simultaneous measurement.Study Type: Prospective.Subjects: Fourteen healthy subjects (35.4 ± 9.3 years, eight males), scanned on Day 1 and Day 7.Field Strength/Sequence: 3 T/3D‐MRF, T₁, T₂, and T_1ρ maps.Assessment: The acquisition of 3D‐MRF cartilage (simultaneous acquisition of T₁, T₂, and T_1ρ maps) were acquired using a dictionary pattern‐matching approach. Conventional cartilage T₁, T₂, and T_1ρ maps were acquired using variable flip angles and a modified 3D‐Turbo‐Flash sequence with different echo and spin‐lock times, respectively, and were fitted using mono‐exponential models. Each sequence was acquired on Day 1 and Day 7 with two scans on each day.Statistical Tests: The mean and SD for cartilage T₁, T₂, and T_1ρ were calculated in five manually segmented regions of interest (ROIs), including lateral femur, lateral tibia, medial femur, medial tibia, and patella cartilages. Intra‐subject and inter‐subject repeatabilities were assessed using coefficient of variation (CV) and intra‐class correlation coefficient (ICC), respectively, on the same day and among different days. Regression and Bland–Altman analysis were performed to compare maps between the conventional and 3D‐MRF sequences.Results: The CV in all ROIs was lower than 7.4%, 8.4%, and 7.5% and the ICC was higher than 0.56, 0.51, and 0.52 for cartilage T₁, T₂, and T_1ρ, respectively. The MRF results had a good agreement with the conventional methods with a linear regression slope >0.61 and R² > 0.59.Conclusion: The 3D‐MRF sequence had high intra‐subject and inter‐subject repeatabilities for simultaneously measuring knee cartilage T₁, T₂, and T_1ρ with good agreement with conventional sequences.Evidence Level: 1Technical Efficacy: Stage 1

Background: There is limited understanding of differences in the composition and structure of ligaments between healthy males and females, and individuals of different ages. Females present higher risk for ligament injuries than males and there are conflicting reports on its cause. This study looks into T_1ρ parameters for an explanation as it relates to proteoglycan, collagen, and water content in these tissues.Purpose: To investigate gender‐related and age‐related differences in T_1ρ parameters in knee joint ligaments in healthy volunteers using a T_1ρ‐prepared zero echo‐time (ZTE)‐based pointwise‐encoding time‐reduction with radial acquisition (T_1ρ‐PETRA) sequence.Study Type: Prospective.Population: The study group consisted of 22 healthy subjects (11 females, ages: 41 ± 18 years, and 11 males, ages: 41 ± 14 years) with no known inflammation, trauma, or pain in the knee joint.Field Strength/Sequence: A T_1ρ‐prepared 3D‐PETRA sequence was used to acquire fat‐suppressed images with varying spin‐lock lengths (TSLs) of the knee joint at 3T.Assessment: Monoexponential, biexponential, and stretched‐exponential 3D‐PETRA‐T_1ρ parameters were measured in the anterior cruciate ligament (ACL), posterior cruciate ligament (PCL), and patellar tendon (PT) by manually drawing ROIs over the entirety of the tissues.Statistical Tests: Mann–Whitney U‐tests were used to compare 3D‐PETRA‐T_1ρ parameters in the ACL, PCL, and PT between males and females. Spearman correlation coefficients were used to determine the association between age and T_1ρ parameters. Statistical significance was defined as P < 0.05.Results: Significant correlations with age were found the three ligaments with most of the measured T_1ρ parameters (r_s = 0.28–0.74) with the exception of the short fraction in the PCL (P = 0.18), and the short relaxation time in the ACL (P = 0.58) and in the PCL (P = 0.14).Data Conclusion: 3D‐PETRA‐T_1ρ can detect age‐related differences in monoexponential, biexponential, and stretched‐exponential T_1ρ parameters in three ligaments of healthy volunteers, which are thought to be related to changes in tissue composition and structure during the aging process.Level of Evidence: 2Technical Efficacy: Stage 1