Search

Artificial intelligence (AI), particularly deep learning, has made enormous strides in medical imaging analysis. In the field of musculoskeletal radiology, deep-learning models are actively being developed for the identification and evaluation of bone fractures. These methods provide numerous benefits to radiologists such as increased diagnostic accuracy and efficiency while also achieving standalone performances comparable or superior to clinician readers. Various algorithms are already commercially available for integration into clinical workflows, with the potential to improve healthcare delivery and shape the future practice of radiology. In this systematic review, we explore the performance of current AI methods in the identification and evaluation of fractures, particularly those in the ankle, wrist, hip, and ribs. We also discuss current commercially available products for fracture detection and provide an overview of the current limitations of this technology and future directions of the field.

Introduction: The direct anterior approach (DAA) and the posterior approach (PA) are 2 common total hip arthroplasty (THA) exposures. This prospective study quantitatively compared changes in periarticular muscle volume after DAA and PA THA., Materials: 19 patients undergoing THA were recruited prospectively from the practices of 3 fellowship-trained hip surgeons. Each surgeon performed a single approach, DAA or PA. Enrolled patients underwent a preoperative MRI of the affected hip and two subsequent postoperative MRIs at around 6 weeks and 6 months after surgery. Clinical evaluations were done by Harris Hip Score at each follow-up interval., Results: MRIs or 10 DAA and 9 PA patients were analysed. Groups did not differ significantly with regard to BMI, age, or preoperative muscle volume. 1 DAA patient suffered a periprosthetic fracture and was excluded from the study. DAA hips showed significant atrophy in the obturator internus (-37.3%) muscle at early follow-up, with persistent atrophy of this muscle at the final follow-up. PA hips showed significant atrophy in the obturator internus (-46.8%) and externus (-16.0%), piriformis (-8.12%), and quadratus femoris muscles (-13.1%) at early follow-up, with persistent atrophy of these muscles at final follow-up. Loss of anterior capsular integrity was present at final follow-up in 2/10 DAA hips while loss of posterior capsular integrity was present in 5/9 PA hips. There was no difference in clinical outcomes., Discussion: This study demonstrates that DAA showed less persistent muscular atrophy than PA. Regardless of surgical approach, a muscle whose tendon is detached from its insertion is likely to demonstrate persistent atrophy 6 months following THA. Although the study was not powered to compare clinical outcomes, it should be noted that no significant difference in patient outcomes was observed.

For accurate diagnosis of interstitial lung disease (ILD), a consensus of radiologic, pathological, and clinical findings is vital. Management of ILD also requires thorough follow-up with computed tomography (CT) studies and lung function tests to assess disease progression, severity, and response to treatment. However, accurate classification of ILD subtypes can be challenging, especially for those not accustomed to reading chest CTs regularly. Dynamic models to predict patient survival rates based on longitudinal data are challenging to create due to disease complexity, variation, and irregular visit intervals. Here, we utilize RadImageNet pretrained models to diagnose five types of ILD with multimodal data and a transformer model to determine a patient's 3-year survival rate. When clinical history and associated CT scans are available, the proposed deep learning system can help clinicians diagnose and classify ILD patients and, importantly, dynamically predict disease progression and prognosis., (© 2023. The Author(s).)

Purpose: To demonstrate the value of pretraining with millions of radiologic images compared with ImageNet photographic images on downstream medical applications when using transfer learning., Materials and Methods: This retrospective study included patients who underwent a radiologic study between 2005 and 2020 at an outpatient imaging facility. Key images and associated labels from the studies were retrospectively extracted from the original study interpretation. These images were used for RadImageNet model training with random weight initiation. The RadImageNet models were compared with ImageNet models using the area under the receiver operating characteristic curve (AUC) for eight classification tasks and using Dice scores for two segmentation problems., Results: The RadImageNet database consists of 1.35 million annotated medical images in 131 872 patients who underwent CT, MRI, and US for musculoskeletal, neurologic, oncologic, gastrointestinal, endocrine, abdominal, and pulmonary pathologic conditions. For transfer learning tasks on small datasets-thyroid nodules (US), breast masses (US), anterior cruciate ligament injuries (MRI), and meniscal tears (MRI)-the RadImageNet models demonstrated a significant advantage ( P < .001) to ImageNet models (9.4%, 4.0%, 4.8%, and 4.5% AUC improvements, respectively). For larger datasets-pneumonia (chest radiography), COVID-19 (CT), SARS-CoV-2 (CT), and intracranial hemorrhage (CT)-the RadImageNet models also illustrated improved AUC ( P < .001) by 1.9%, 6.1%, 1.7%, and 0.9%, respectively. Additionally, lesion localizations of the RadImageNet models were improved by 64.6% and 16.4% on thyroid and breast US datasets, respectively., Conclusion: RadImageNet pretrained models demonstrated better interpretability compared with ImageNet models, especially for smaller radiologic datasets. Keywords: CT, MR Imaging, US, Head/Neck, Thorax, Brain/Brain Stem, Evidence-based Medicine, Computer Applications-General (Informatics) Supplemental material is available for this article. Published under a CC BY 4.0 license.See also the commentary by Cadrin-Chênevert in this issue., Competing Interests: Disclosures of conflicts of interest: X.M. Member of Radiology: Artificial Intelligence trainee editorial board. Z.L. No relevant relationships. P.M.R. No relevant relationships. B.M. Consulting fees from Coridea. M.H. Portable knee MRI unit lent by Hyperfine to author’s institution, free of charge, for research purposes. A.D. No relevant relationships. A.J. No relevant relationships. C.C. No relevant relationships. K.E.L. Medical student research grant from Radiological Society of North America, payment made to NYU Langone; medical student scholarship from American College of Radiology, payment made to author. T.Y. No relevant relationships. Y.W. Grant support provided to author’s institution through NSF DMS-1752709 and NSF DMS-1720489 by the National Science Foundation. H.G. No relevant relationships. T.D. Radiologic data for the study was provided by RadImageNet and author is managing partner of RadImageNet. Z.A.F. Grants from Daiichi Sankyo, Amgen, Bristol Myers Squibb, and Siemens Healthineers; personal fees from Alexion, GlaxoSmithKline, and Trained Therapeutix Discovery; patents with Trained Therapeutix Discovery. Y.Y. Mount Sinai Seed Fund., (© 2022 by the Radiological Society of North America, Inc.)

For diagnosis of coronavirus disease 2019 (COVID-19), a SARS-CoV-2 virus-specific reverse transcriptase polymerase chain reaction (RT-PCR) test is routinely used. However, this test can take up to 2 d to complete, serial testing may be required to rule out the possibility of false negative results and there is currently a shortage of RT-PCR test kits, underscoring the urgent need for alternative methods for rapid and accurate diagnosis of patients with COVID-19. Chest computed tomography (CT) is a valuable component in the evaluation of patients with suspected SARS-CoV-2 infection. Nevertheless, CT alone may have limited negative predictive value for ruling out SARS-CoV-2 infection, as some patients may have normal radiological findings at early stages of the disease. In this study, we used artificial intelligence (AI) algorithms to integrate chest CT findings with clinical symptoms, exposure history and laboratory testing to rapidly diagnose patients who are positive for COVID-19. Among a total of 905 patients tested by real-time RT-PCR assay and next-generation sequencing RT-PCR, 419 (46.3%) tested positive for SARS-CoV-2. In a test set of 279 patients, the AI system achieved an area under the curve of 0.92 and had equal sensitivity as compared to a senior thoracic radiologist. The AI system also improved the detection of patients who were positive for COVID-19 via RT-PCR who presented with normal CT scans, correctly identifying 17 of 25 (68%) patients, whereas radiologists classified all of these patients as COVID-19 negative. When CT scans and associated clinical history are available, the proposed AI system can help to rapidly diagnose COVID-19 patients.

Background: Ultrasound (US) is the most commonly used radiologic modality to identify and characterize thyroid nodules. Many nodules subsequently undergo fine needle aspiration to further characterize the nodule and determine appropriate treatment. The fine needle aspirate is most commonly classified using the Bethesda System for Reporting Thyroid Cytology (TBSRTC). It can sometimes be difficult to differentiate Bethesda class III lesions (atypia of undetermined significance/follicular lesion of undetermined significance) from Bethesda class IV, V and VI (malignant nodules). However, differentiation is important as clinical management differs between the two groups. The purpose of this study was to introduce machine learning methods to help radiologists differentiate Bethesda class III from Bethesda class VI, V and VI lesions., Methods: The authors collected 467 thyroid nodules with cytopathology results. US features were summarized using the 2017 ACR (American College of Radiology) Thyroid Imaging Reporting And Data System (TIRADS). Machine learning models [logistic regression, gradient boost, support vector machine (SVM), random forest and deep neural networks (DNN)] were created to classify Bethesda class III vs class IV/V/VI., Results: DNN outperformed other machine learning classifiers and obtained the highest accuracy and specificity to classify thyroid nodules as either Bethesda III or IV/V/VI nodules using multiple US features., Conclusions: Machine learning/deep learning approaches could help differentiate Bethesda III nodules from IV/V/VI using US features which may benefit treatment decisions., Competing Interests: Conflicts of Interest: The authors have no conflicts of interest to declare.

Competing Interests: Conflicts of Interest: The authors have no conflicts of interest to declare.

Background: The purpose of this study was to develop a predictive model that could accurately predict the malignancy of the pulmonary ground-glass nodules (GGNs) and the invasiveness of the malignant GGNs., Methods: The authors built two binary classification models that could predict the malignancy of the pulmonary GGNs and the invasiveness of the malignant GGNs., Results: Results of our developed model showed random forest could achieve 95.1% accuracy to predict the malignancy of GGNs and 83.0% accuracy to predict the invasiveness of the malignant GGNs., Conclusions: The malignancy and invasiveness of pulmonary GGNs could be predicted by random forest., Competing Interests: Conflicts of Interest: The authors have no conflicts of interest to declare.

Background: Autologous osteochondral transplantation (AOT) has an inherent risk of donor site morbidity (DSM). The reported rates of DSM vary from 0% to 50%, with few studies reporting clinical or imaging outcomes at the donor site as a primary outcome and even fewer report these outcomes when a biosynthetic plug backfill is employed. Although TruFit (Smith & Nephew, Andover, MA) plugs have been removed from the market for regulatory purposes, biphasic plugs (including TruFit plugs) have been used for several years and the evaluation of these is therefore pertinent., Methods: Thirty-nine patients who underwent forty AOT procedures of the talus, with the donor graft being taken from the ipsilateral knee, were included. Postoperative magnetic resonance imaging (MRI) was used to assess the donor site graded with magnetic resonance observation of cartilage repair tissue (MOCART) scoring. Lysholm scores were collected preoperatively, at the time of magnetic resonance imaging (MRI), and again at 24 months and at final follow-up to assess clinical outcomes. Statistical analysis was performed to establish if there was any correlation between MRI assessment of the donor site and clinical outcomes. The mean patient age was 36.2 ± 15.7 years with a mean follow-up of 41.8 ± 16.7 months., Results: All patient donor site defects were filled with OBI TruFit biphasic plugs. DSM was encountered in 12.5% of the patient cohort at 24 months, and in these patients, the Lysholm score was a mean 87.2 ± 5.0. At final follow-up, DSM was reduced to 5%. Lysholm scores for the entire cohort were 98.4 ± 4.6 and 99.4 ± 3.1 at 24 months and final follow-up, respectively. MRI of the donor sites were taken at an average of 18.1 ± 13.5 (range, 3-48) months postoperatively and the mean MOCART score was 60.0 ± 13.5. No correlation was found between the MOCART score and Lysholm outcomes at the donor knee (P = .43, r = 0.13)., Conclusion: Low incidence of DSM and good functional outcomes were achieved with AOT. Additionally, MRI findings did not predict clinical outcomes in our study., Level of Evidence: Level IV, retrospective case series., (© The Author(s) 2016.)

Background: The etiology of lateral hindfoot pain in flatfoot deformity can be difficult to assess on plain radiographs. We hypothesized that multiplanar measurements obtained in a fully upright, weightbearing position would reliably demonstrate an increase in lateral hindfoot arthrosis and/or impingement in a cohort of flatfoot patients with lateral hindfoot pain compared to a cohort without pain., Materials and Methods: Ten consecutive patients with flexible flatfoot deformity and lateral hindfoot pain (mean age, 55.5 +/- 13.9) were compared to 10 consecutive patients with flexible deformity but no pain (mean age, 61.0 +/- 8.6). Conventional radiographs and weightbearing multiplanar images were performed on all patients before surgical reconstruction. Flatfoot, hindfoot impingement, and arthrosis parameters were interpreted in a blinded fashion by two musculoskeletal radiologists. Interrater reliability was determined with intraclass correlation coefficients (ICC). All parameters were compared between groups with a Wilcoxon rank sum test (p < 0.05)., Results: A significant increase in posterior facet subtalar arthrosis (p = 0.006) and combined anterior and posterior facet subtalar arthrosis (p = 0.022) was evident in the pain group. Calcaneofibular impingement and calcaneocuboid arthritis were increased in the pain group, but did not reach significance (p = 0.057 and p = 0.067 respectively). The multiplanar imaging parameters demonstrated good (ICC = 0.60 to 0.73) to excellent (ICC >or= 0.74) reliability for most impingement and arthrosis parameters and for many of the standard flatfoot parameters., Conclusion: The results indicate that weightbearing, multiplanar imaging provides a reliable means of assessing lateral pain in patients with flexible flatfoot deformity.

Objective: To test the (null) hypotheses that the reliability of unipedal balance is unaffected by the attenuation of visual velocity feedback and that, relative to baseline performance, deterioration of balance success rates from attenuated visual velocity feedback will not differ between groups of young men and older women, and the presence (or absence) of a vertical foreground object will not affect balance success rates., Design: Single blind, single case study., Setting: University research laboratory., Participants: Two volunteer samples: 26 healthy young men (mean age, 20.0yrs; SD, 1.6); 23 healthy older women (mean age, 64.9 yrs; SD, 7.8)., Main Outcome Measure: Normalized success rates in unipedal balance task., Methods: Subjects were asked to transfer to and maintain unipedal stance for 5 seconds in a task near the limit of their balance capabilities. Subjects completed 64 trials: 54 trials of three experimental visual scenes in blocked randomized sequences of 18 trials and 10 trials in a normal visual environment. The experimental scenes included two that provided strong velocity/weak position feedback, one of which had a vertical foreground object (SVWP+) and one without (SVWP-), and one scene providing weak velocity/strong position (WVSP) feedback. Subjects' success rates in the experimental environments were normalized by the success rate in the normal environment in order to allow comparisons between subjects using a mixed model repeated measures analysis of variance., Results: The normalized success rate was significantly greater in SVWP+ than in WVSP (p = .0001) and SVWP- (p = .013). Visual feedback significantly affected the normalized unipedal balance success rates (p = .001); neither the group effect nor the group X visual environment interaction was significant (p = .9362 and p = .5634, respectively). Normalized success rates did not differ significantly between the young men and older women in any visual environment., Conclusions: Near the limit of the young men's or older women's balance capability, the reliability of transfer to unipedal balance was adversely affected by visual environments offering attenuated visual velocity feedback cues and those devoid of vertical foreground objects.