Checklist for Reproducibility of Deep Learning in Medical Imaging.

The application of deep learning (DL) in medicine introduces transformative tools with the potential to enhance prognosis, diagnosis, and treatment planning. However, ensuring transparent documentation is essential for researchers to enhance reproducibility and refine techniques. Our study addresses the unique challenges presented by DL in medical imaging by developing a comprehensive checklist using the Delphi method to enhance reproducibility and reliability in this dynamic field. We compiled a preliminary checklist based on a comprehensive review of existing checklists and relevant literature. A panel of 11 experts in medical imaging and DL assessed these items using Likert scales, with two survey rounds to refine responses and gauge consensus. We also employed the content validity ratio with a cutoff of 0.59 to determine item face and content validity. Round 1 included a 27-item questionnaire, with 12 items demonstrating high consensus for face and content validity that were then left out of round 2. Round 2 involved refining the checklist, resulting in an additional 17 items. In the last round, 3 items were deemed non-essential or infeasible, while 2 newly suggested items received unanimous agreement for inclusion, resulting in a final 26-item DL model reporting checklist derived from the Delphi process. The 26-item checklist facilitates the reproducible reporting of DL tools and enables scientists to replicate the study's results.

Robust Multi-View Fracture Detection in the Presence of Other Abnormalities Using HAMIL-Net.

INTRODUCTION: Foot and ankle fractures are the most common military health problem. Automated diagnosis can save time and personnel. It is crucial to distinguish fractures not only from normal healthy cases, but also robust against the presence of other orthopedic pathologies. Artificial intelligence (AI) deep learning has been shown to be promising. Previously, we have developed HAMIL-Net to automatically detect orthopedic injuries for upper extremity injuries. In this research, we investigated the performance of HAMIL-Net for detecting foot and ankle fractures in the presence of other abnormalities. MATERIALS AND METHODS: HAMIL-Net is a novel deep neural network consisting of a hierarchical attention layer followed by a multiple-instance learning layer. The design allowed it to deal with imaging studies with multiple views. We used 148K musculoskeletal imaging studies for 51K Veterans at VA San Diego in the past 20 years to create datasets for this research. We annotated each study by a semi-automated pipeline leveraging radiology reports written by board-certified radiologists and extracting findings with a natural language processing tool and manually validated the annotations. RESULTS: HAMIL-Net can be trained with study-level, multiple-view examples, and detect foot and ankle fractures with a 0.87 area under the receiver operational curve, but the performance dropped when tested by cases including other abnormalities. By integrating a fracture specialized model with one that detecting a broad range of abnormalities, HAMIL-Net's accuracy of detecting any abnormality improved from 0.53 to 0.77 and F-score from 0.46 to 0.86. We also reported HAMIL-Net's performance under different study types including for young (age 18-35) patients. CONCLUSIONS: Automated fracture detection is promising but to be deployed in clinical use, presence of other abnormalities must be considered to deliver its full benefit. Our results with HAMIL-Net showed that considering other abnormalities improved fracture detection and allowed for incidental findings of other musculoskeletal abnormalities pertinent or superimposed on fractures.

Reproducibility of Deep Learning Algorithms Developed for Medical Imaging Analysis: A Systematic Review.

Since 2000, there have been more than 8000 publications on radiology artificial intelligence (AI). AI breakthroughs allow complex tasks to be automated and even performed beyond human capabilities. However, the lack of details on the methods and algorithm code undercuts its scientific value. Many science subfields have recently faced a reproducibility crisis, eroding trust in processes and results, and influencing the rise in retractions of scientific papers. For the same reasons, conducting research in deep learning (DL) also requires reproducibility. Although several valuable manuscript checklists for AI in medical imaging exist, they are not focused specifically on reproducibility. In this study, we conducted a systematic review of recently published papers in the field of DL to evaluate if the description of their methodology could allow the reproducibility of their findings. We focused on the Journal of Digital Imaging (JDI), a specialized journal that publishes papers on AI and medical imaging. We used the keyword "Deep Learning" and collected the articles published between January 2020 and January 2022. We screened all the articles and included the ones which reported the development of a DL tool in medical imaging. We extracted the reported details about the dataset, data handling steps, data splitting, model details, and performance metrics of each included article. We found 148 articles. Eighty were included after screening for articles that reported developing a DL model for medical image analysis. Five studies have made their code publicly available, and 35 studies have utilized publicly available datasets. We provided figures to show the ratio and absolute count of reported items from included studies. According to our cross-sectional study, in JDI publications on DL in medical imaging, authors infrequently report the key elements of their study to make it reproducible.

RadBERT: Adapting Transformer-based Language Models to Radiology.

Purpose: To investigate if tailoring a transformer-based language model to radiology is beneficial for radiology natural language processing (NLP) applications. Materials and Methods: This retrospective study presents a family of bidirectional encoder representations from transformers (BERT)-based language models adapted for radiology, named RadBERT. Transformers were pretrained with either 2.16 or 4.42 million radiology reports from U.S. Department of Veterans Affairs health care systems nationwide on top of four different initializations (BERT-base, Clinical-BERT, robustly optimized BERT pretraining approach [RoBERTa], and BioMed-RoBERTa) to create six variants of RadBERT. Each variant was fine-tuned for three representative NLP tasks in radiology: (a) abnormal sentence classification: models classified sentences in radiology reports as reporting abnormal or normal findings; (b) report coding: models assigned a diagnostic code to a given radiology report for five coding systems; and (c) report summarization: given the findings section of a radiology report, models selected key sentences that summarized the findings. Model performance was compared by bootstrap resampling with five intensively studied transformer language models as baselines: BERT-base, BioBERT, Clinical-BERT, BlueBERT, and BioMed-RoBERTa. Results: For abnormal sentence classification, all models performed well (accuracies above 97.5 and F1 scores above 95.0). RadBERT variants achieved significantly higher scores than corresponding baselines when given only 10% or less of 12 458 annotated training sentences. For report coding, all variants outperformed baselines significantly for all five coding systems. The variant RadBERT-BioMed-RoBERTa performed the best among all models for report summarization, achieving a Recall-Oriented Understudy for Gisting Evaluation-1 score of 16.18 compared with 15.27 by the corresponding baseline (BioMed-RoBERTa, P < .004). Conclusion: Transformer-based language models tailored to radiology had improved performance of radiology NLP tasks compared with baseline transformer language models.Keywords: Translation, Unsupervised Learning, Transfer Learning, Neural Networks, Informatics Supplemental material is available for this article. © RSNA, 2022See also commentary by Wiggins and Tejani in this issue.

Federated learning for predicting clinical outcomes in patients with COVID-19.

Federated learning (FL) is a method used for training artificial intelligence models with data from multiple sources while maintaining data anonymity, thus removing many barriers to data sharing. Here we used data from 20 institutes across the globe to train a FL model, called EXAM (electronic medical record (EMR) chest X-ray AI model), that predicts the future oxygen requirements of symptomatic patients with COVID-19 using inputs of vital signs, laboratory data and chest X-rays. EXAM achieved an average area under the curve (AUC) >0.92 for predicting outcomes at 24 and 72 h from the time of initial presentation to the emergency room, and it provided 16% improvement in average AUC measured across all participating sites and an average increase in generalizability of 38% when compared with models trained at a single site using that site's data. For prediction of mechanical ventilation treatment or death at 24 h at the largest independent test site, EXAM achieved a sensitivity of 0.950 and specificity of 0.882. In this study, FL facilitated rapid data science collaboration without data exchange and generated a model that generalized across heterogeneous, unharmonized datasets for prediction of clinical outcomes in patients with COVID-19, setting the stage for the broader use of FL in healthcare.

Federated Learning used for predicting outcomes in SARS-COV-2 patients.

'Federated Learning' (FL) is a method to train Artificial Intelligence (AI) models with data from multiple sources while maintaining anonymity of the data thus removing many barriers to data sharing. During the SARS-COV-2 pandemic, 20 institutes collaborated on a healthcare FL study to predict future oxygen requirements of infected patients using inputs of vital signs, laboratory data, and chest x-rays, constituting the "EXAM" (EMR CXR AI Model) model. EXAM achieved an average Area Under the Curve (AUC) of over 0.92, an average improvement of 16%, and a 38% increase in generalisability over local models. The FL paradigm was successfully applied to facilitate a rapid data science collaboration without data exchange, resulting in a model that generalised across heterogeneous, unharmonized datasets. This provided the broader healthcare community with a validated model to respond to COVID-19 challenges, as well as set the stage for broader use of FL in healthcare.

Identifying biopsychosocial factors that impact decompressive laminectomy outcomes in veterans with lumbar spinal stenosis: a prospective cohort study.

ABSTRACT: One in 3 patients with lumbar spinal stenosis undergoing decompressive laminectomy (DL) to alleviate neurogenic claudication do not experience substantial improvement. This prospective cohort study conducted in 193 Veterans aimed to identify key spinal and extraspinal factors that may contribute to a favorable DL outcome. Biopsychosocial factors evaluated pre-DL and 1 year post-DL were hip osteoarthritis, imaging-rated severity of spinal stenosis, scoliosis/kyphosis, leg length discrepancy, comorbidity, fibromyalgia, depression, anxiety, pain coping, social support, pain self-efficacy, sleep, opioid and nonopioid pain medications, smoking, and other substance use. The Brigham Spinal Stenosis (BSS) questionnaire was the main outcome. Brigham Spinal Stenosis scales (symptom severity, physical function [PF], and satisfaction [SAT]) were dichotomized as SAT < 2.42, symptom severity improvement ≥ 0.46, and PF improvement ≥ 0.42, and analyzed using logistic regression. Sixty-two percent improved in 2 of 3 BSS scales (ie, success). Baseline characteristics associated with an increased odds of success were-worse BSS PF (odds ratio [OR] 1.24 [1.08-1.42]), greater self-efficacy for PF (OR 1.30 [1.08-1.58]), lower self-efficacy for pain management (OR 0.80 [0.68-0.94]), less apparent leg length discrepancy (OR 0.71 [0.56-0.91]), greater self-reported alcohol problems (OR 1.53 [1.07-2.18]), greater treatment credibility (OR 1.31 [1.07-1.59]), and moderate or severe magnetic resonance imaging-identified central canal stenosis (OR 3.52 [1.06-11.6]) moderate, OR 5.76 [1.83-18.1] severe). Using opioids was associated with lower odds of significant functional improvement (OR 0.46 [0.23-0.93]). All P < 0.05. Key modifiable factors associated with DL success-self-efficacy, apparent leg length inequality, and opioids-require further investigation and evaluation of the impact of their treatment on DL outcomes.

Patterns of cartilage degeneration in knees with medial tibiofemoral offset.

OBJECTIVE: To determine if radiographic medial tibiofemoral offset (MTFO) is associated with: (1) magnetic resonance imaging (MRI) pathology of cartilage, meniscus, and ligament; and (2) a distinct pattern of lateral cartilage degeneration on MRI. MATERIALS AND METHODS: Three hundred consecutive adult knee MRIs with anteroposterior (AP) radiographs were retrospectively reviewed, and 145 studies were included. MTFO was defined as a medial extension of the medial femoral condyle beyond the articular surface of the medial tibial plateau on weight-bearing AP radiographs. The patients were then divided into the MTFO (n = 61) or no-offset (n = 84) groups. On MRI data obtained on a 1.5-Tesla system, articular cartilage of the femoral condyle and tibial plateau were graded using a modified Outerbridge classification (36 sub-regions similar to whole-organ MRI Score (WORMS) system). In addition, MR pathology of the ACL, MCL, LCL, medial and lateral menisci, were determined. RESULTS: Significantly increased (ANOVA p < 0.007) MR grade of the ligaments, menisci, and cartilage in the MTFO group (ranging from 0.3 to 2.5) compared to the control group (0.2 to 1.1). Color maps of the cartilage grades suggested a marked difference in both severity of degeneration and regional variations between the groups. MTFO group exhibited focally increased cartilage grades in the central, non-weight regions of lateral compartment (region p = 0.07 to 0.12, interaction p = 0.05 to 0.1). CONCLUSIONS: MTFO is associated with overall degeneration of the knee and features a distinct lateral cartilage degeneration pattern, which may reflect non-physiologic contact of the cartilage between the lateral tibial eminence and lateral central femoral condyle.

Cost accounting for the radiologist.

OBJECTIVE: Cost accounting is the branch of managerial accounting that deals with the analysis of the costs of a product or service. This article reviews methods of classifying and allocating costs and relationships among costs, volume, and revenues. CONCLUSION: Radiology practices need to know the cost of a procedure or service to determine the selling price of a product, bid on contracts, analyze profitability, and facilitate cost control and cost reduction.

Degenerative changes in the ligamentum teres of the hip: cadaveric study with magnetic resonance arthrography, anatomical inspection, and histologic examination.

OBJECTIVE: Review and classify the pathologic abnormalities of the ligamentum teres in degenerative hip joints. MATERIAL AND METHODS: Eleven cadaveric hip joints were examined with magnetic resonance arthrography and were then sectioned. The appearance of the ligamentum teres and its attachment seen by inspection of the anatomical sections was correlated with findings seen at magnetic resonance imaging. Histologic evaluation was done. RESULTS: Magnetic resonance arthrographic and histologic findings showed a spectrum of ligamentum teres degeneration. The thickest ligamentum teres revealed mucoid and fibromatous degeneration with microscopic tear. Ligamentum teres with intermediate thickness revealed fatty replacement with and without fibromatous degeneration, fibromatous degeneration with and without mucoid degeneration, and eosinophilic change. The thinnest ligamentum teres was near-complete disruption of the ligament. CONCLUSIONS: The ligamentum teres revealed degenerative changes that were similar histologically to degenerative changes in tendons. Magnetic resonance arthrography provided a sensitive technique for demonstration of the ligamentum teres and various patterns of degeneration.

Magnetic resonance imaging of the hip.

Standard magnetic resonance imaging (MRI) as well as MR arthrography (MRA) have been important diagnostic tools to assess for a spectrum of clinical presentations related to the hip. MRA has allowed the radiologist to closely examine intracapsular structures such as the acetabular labrum. In this article, we provide a general review of soft tissue and osseous anatomy of hips, especially focusing on the MR appearances of the acetabular labrum and the osseous morphology of the greater trochanter and ischial tuberosity with their muscle and tendon attachments. In addition, current topics in recent literature will be discussed such as femoroacetabular impingement (FAI) and rotator cuff tears of the hip.

Musculoskeletal radiology curriculum for radiology residents.

This article outlines a curriculum for resident education in musculoskeletal radiology that addresses the current requirements for assessment of the general competencies as set forth by the Accreditation Council for Graduate Medical Education.

Informatics in radiology: use of the MIRC DICOM service for clinical trials to automatically create teaching file cases from PACS.

The Medical Imaging Resource Center (MIRC) software of the Radiological Society of North America is used to share teaching files and clinical trial data. By using the MIRC service for clinical trials, teaching file cases can be automatically generated directly from picture archiving and communication systems (PACS). The advantage of using the clinical trials service over the authoring service is the need for minimal user intervention, because data already present in the Digital Imaging and Communications in Medicine (DICOM) header can be automatically inserted into teaching file cases. When reading a case on the PACS workstation, the radiologist needs only to select the images for the teaching file case, write the diagnosis in the "Study Comments" box (usually used for "wet" readings), and send the case to the MIRC server directly from the workstation. The MIRC server automatically makes the images anonymous and fills in the modified teaching file template by using the data in the DICOM header (patient age, patient sex, clinical history, diagnosis, imaging modality, organ system), thus creating a simple teaching file. If desired, the teaching file case can be edited at a later time. By making minor modifications to the MIRC DICOM clinical trials service, it is possible to create teaching file cases with minimal effort. Supplemental material available at radiographics.rsnajnls.org/cgi/content/full/27/1/269/DC1.

The advanced imaging of gouty tophi.

Gout is a metabolic disorder in which there is hyperuricemia caused by an increase in production or a decrease in excretion of uric acid. Long-lasting hyperuricemia causes the deposition of monosodium urate crystals in the joints and soft tissues, triggering gouty arthritis and, if not properly treated, the formation of gouty tophi. The diagnosis of gout is usually based on clinical presentation and laboratory examinations, long before any abnormality can be demonstrated with imaging. Radiography is the primary imaging modality used in the initial evaluation of gouty arthritis. Ultrasonography, CT, MRI, and nuclear medicine are seldom necessary. Occasionally a tophus has an unusual presentation and simulates neoplasm or infection prompting the utilization of cross-sectional imaging for further evaluation and surgical planning. Cross-sectional imaging is also used in areas that are difficult to visualize on radiographs such as spine, sacroiliac joints, and soft tissues.

MR imaging of the finger tendons: normal anatomy and commonly encountered pathology.

MR imaging has emerged as a powerful tool in the evaluation of soft tissue structures such as the tendons of the hand and finger due to its excellent soft tissue contrast and multiplanar imaging capabilities. In the finger and hand, perhaps more than in any other location in the body, a detailed and intimate understanding of anatomy is crucial for lesion localization, directing clinical management and predicting long-term prognosis. These issues are of paramount importance to both the clinician and imager, both faced with the challenge of the complex anatomy and pathology associated with these delicate structures. The anatomy of the finger including intrinsic and extrinsic muscles, retinacular structures, and tendons will be discussed. The MR imaging features of common lesions of the tendons of the hand and finger will be reviewed.

Magnetic resonance imaging of elbow instability.

Traumatic injury of the elbow, both in the acute setting as well as in the setting of chronic repetitive microtrauma, is quite common. Specifically, the frequency of elbow dislocation is second only to that of the shoulder. Unlike the shoulder, however, the elbow is an inherently stable articulation, considered one of the most congruous joints in the body. This manuscript discussed imaging techniques for elbow evaluation in the setting of instability and reviews the osseous anatomy, as well as the anatomy of the soft tissue stabilizers of the elbow. Anatomic variants, potential pitfalls in imaging diagnosis, and the imaging findings associated with various types of elbow instability are considered.

Calcific tendinosis and periarthritis: classic magnetic resonance imaging appearance and associated findings.

Calcification within tendon and periarticular soft tissues with or without a history of pain is commonly encountered in the interpretation of musculoskeletal MR imaging (MRI) studies. This article presents a pictorial illustration of both the classic and atypical MR imaging (MRI) features of soft tissue calcifications in the musculoskeletal system. Familiarity with this topic can help to avoid pitfalls in imaging diagnosis.

Advanced imaging of gout.

Gout is a metabolic disorder in which there is either an increase in production or a decrease in excretion of uric acid leading to hyperuricemia. Long-lasting hyperuricemia causes the deposition of monosodium urate crystals in the joints and soft tissues triggering gouty arthritis and, if not properly treated, the formation of gouty tophi. Characteristic of gout are well-defined, punched-out erosion with overhanging edges, with preservation of the joint space, lack of periarticular osteopenia, asymmetrical involvement, soft tissue nodules, and intraosseous calcifications. On magnetic resonance imaging, tophi usually have low signal intensity on both TI- and T2-weighted images and a variable enhancement pattern.

MR imaging of soft-tissue masses of the foot.

As with other parts of the musculoskeletal system, the soft tissues of the foot can be affected by a wide variety of pathologic entities including trauma, congenital abnormalities, infections, and neoplastic disorders. While plain radiographs are usually the initial examination for evaluation of pathology, magnetic resonance imaging (MRI) is critical to evaluate for abnormalities within the ligaments, tendons, and other nonosseous structures within the foot. The constellation of clinical and MRI findings often allows a relatively specific diagnosis to be rendered. This article discusses both benign and malignant processes within the soft tissues of the foot and presents their characteristic imaging findings with MRI.

Interactive Atlas of Sign in Musculoskeletal Radiology

It presents images and information related to common and not so common signs used in Musculoskeletal Radiology, which can be reviewed by sign name (from Absent bow tie sign to Vacuum phenomenon), pathological diagnosis or locations (from Ankle to Wrist). The signs are illustrated with radiographs and diagrams. The references are linked to PubMed, allowing access to the abstract, if available.