Radiology-Artificial Intelligence最新文献

Purpose To develop and evaluate machine learning and deep learning-based models for automated protocoling of emergency brain MRI scans based on clinical referral text. Materials and Methods In this single-institution, retrospective study of 1953 emergency brain MRI referrals from January 2016 to January 2019, two neuroradiologists labeled the imaging protocol and use of contrast agent as the reference standard. Three machine learning algorithms (naive Bayes, support vector machine, and XGBoost) and two pretrained deep learning models (Finnish bidirectional encoder representations from transformers [BERT] and generative pretrained transformer [GPT]-3.5 [GPT-3.5 Turbo; Open AI]) were developed to predict the MRI protocol and need for a contrast agent. Each model was trained with three datasets (100% of training data, 50% of training data, and 50% plus augmented training data). Prediction accuracy was assessed with a test set. Results The GPT-3.5 models trained with 100% of the training data performed best in both tasks, achieving an accuracy of 84% (95% CI: 80, 88) for the correct protocol and 91% (95% CI: 88, 94) for the contrast agent. BERT had an accuracy of 78% (95% CI: 74, 82) for the protocol and 89% (95% CI: 86, 92) for the contrast agent. The best machine learning model in the protocol task was XGBoost (accuracy, 78%; 95% CI: 73, 82), and the best machine learning models in the contrast agent task were support vector machine and XGBoost (accuracy, 88%; 95% CI: 84, 91 for both). The accuracies of two nonneuroradiologists were 80%-83% in the protocol task and 89%-91% in the contrast medium task. Conclusion Machine learning and deep learning models demonstrated high performance in automatic protocoling of emergency brain MRI scans based on text from clinical referrals. Keywords: Natural Language Processing, Automatic Protocoling, Deep Learning, Machine Learning, Emergency Brain MRI Supplemental material is available for this article. Published under a CC BY 4.0 license. See also commentary by Strotzer in this issue.

Purpose To test a commercial artificial intelligence (AI) system for breast cancer detection at the BC Cancer Breast Screening Program. Materials and Methods In this retrospective study of 136 700 female individuals (mean age, 58.8 years ± 9.4 [SD]; median, 59.0 years; IQR = 14.0) who underwent digital mammography screening in British Columbia, Canada, between February 2019 and January 2020, breast cancer detection performance of a commercial AI algorithm was stratified by demographic, clinical, and imaging features and evaluated using the area under the receiver operating characteristic curve (AUC), and AI performance was compared with radiologists, using sensitivity and specificity. Results At 1-year follow-up, the AUC of the AI algorithm was 0.93 (95% CI: 0.92, 0.94) for breast cancer detection. Statistically significant differences were found for mammograms across radiologist-assigned Breast Imaging Reporting and Data System breast densities: category A, AUC of 0.96 (95% CI: 0.94, 0.99); category B, AUC of 0.94 (95% CI: 0.92, 0.95); category C, AUC of 0.93 (95% CI: 0.91, 0.95), and category D, AUC of 0.84 (95% CI: 0.76, 0.91) (A_AUC > D_AUC, P = .002; B_AUC > D_AUC, P = .009; C_AUC > D_AUC, P = .02). The AI showed higher performance for mammograms with architectural distortion (0.96 [95% CI: 0.94, 0.98]) versus without (0.92 [95% CI: 0.90, 0.93], P = .003) and lower performance for mammograms with calcification (0.87 [95% CI: 0.85, 0.90]) versus without (0.92 [95% CI: 0.91, 0.94], P < .001). Sensitivity of radiologists (92.6% ± 1.0) exceeded the AI algorithm (89.4% ± 1.1, P = .01), but there was no evidence of difference at 2-year follow-up (83.5% ± 1.2 vs 84.3% ± 1.2, P = .69). Conclusion The tested commercial AI algorithm is generalizable for a large external breast cancer screening cohort from Canada but showed different performance for some subgroups, including those with architectural distortion or calcification in the image. Keywords: Mammography, QA/QC, Screening, Technology Assessment, Screening Mammography, Artificial Intelligence, Breast Cancer, Model Testing, Bias and Fairness Supplemental material is available for this article. Published under a CC BY 4.0 license. See also commentary by Milch and Lee in this issue.

Purpose To evaluate the predictive value of deep learning (DL)-based coronary artery disease (CAD) extent analysis for major adverse cardiac events (MACEs) in patients with acute chest pain presenting to the emergency department (ED). Materials and Methods This retrospective multicenter observational study included consecutive patients with acute chest pain who underwent coronary CT angiography (CCTA) at three institutional EDs from January 2018 to December 2022. Patients were classified as having no CAD, nonobstructive CAD, or obstructive CAD using a DL model. The primary outcome was MACEs during follow-up, defined as a composite of cardiac death, nonfatal myocardial infarction, and hospitalization for unstable angina. Cox proportional hazards regression models were used to evaluate the predictors of MACEs. Results The study included 408 patients (224 male; mean age, 59.4 years ± 14.6 [SD]). The DL model classified 162 (39.7%) patients as having no CAD, 94 (23%) as having nonobstructive CAD, and 152 (37.3%) as having obstructive CAD. Sixty-three (15.4%) patients experienced MACEs during follow-up. Patients with MACEs had a higher prevalence of obstructive CAD than those without (P < .001). In the multivariate analysis of model 1 (clinical risk factors), dyslipidemia (hazard ratio [HR], 2.15) and elevated troponin T levels (HR, 2.13) were predictive of MACEs (all P < .05). In model 2 (clinical risk factors plus DL-based CAD extent), obstructive CAD detected by the DL model was the most significant independent predictor of MACEs (HR, 88.07; P < .001). Harrell C statistic showed that DL-based CAD extent enhanced the risk stratification beyond clinical risk factors (Harrell C statistics: 0.94 vs 0.80, P < .001). Conclusion DL-based detection of obstructive CAD demonstrated stronger predictive value than clinical risk factors for MACEs in patients with acute chest pain presenting to the ED. Keywords: Cardiac, CT-Angiography, Outcomes Analysis © RSNA, 2025 See also commentary by Reddy in this issue.

Purpose To assess the prognostic value of an open-source deep learning-based chest radiographs algorithm, CXR-Lung-Risk, for stratifying respiratory disease mortality risk among an Asian health screening population using baseline and follow-up chest radiographs. Materials and Methods This single-center, retrospective study analyzed chest radiographs from individuals who underwent health screenings between January 2004 and June 2018. The CXR-Lung-Risk scores from baseline chest radiographs were externally tested for predicting mortality due to lung disease or lung cancer, using competing risk analysis, with adjustments made for clinical factors. The additional value of these risk scores beyond clinical factors was evaluated using the likelihood ratio test. An exploratory analysis was conducted on the CXR-Lung-Risk trajectory over a 3-year follow-up period for individuals in the highest quartile of baseline respiratory disease mortality risk, using a time-series clustering algorithm. Results Among 36 924 individuals (median age, 58 years [IQR, 53-62 years]; 22 352 male), 264 individuals (0.7%) died of respiratory illness, over a median follow-up period of 11.0 years (IQR, 7.8-12.7 years). CXR-Lung-Risk predicted respiratory disease mortality (adjusted hazard ratio [HR] per 5 years: 2.01; 95% CI: 1.76, 2.39; P < .001), offering a prognostic improvement over clinical factors (P < .001). The trajectory analysis identified a subgroup with a continuous increase in CXR-Lung-Risk score, which was associated with poorer outcomes (adjusted HR for respiratory disease mortality: 3.26; 95% CI: 1.20, 8.81; P = .02) compared with the subgroup with a continuous decrease in CXR-Lung-Risk score. Conclusion The open-source CXR-Lung-Risk model predicted respiratory disease mortality in an Asian cohort, enabling a two-layer risk stratification approach through an exploratory longitudinal analysis of baseline and follow-up chest radiographs. Keywords: Conventional Radiography, Thorax, Lung, Mediastinum, Heart, Outcomes Analysis Supplemental material is available for this article. © RSNA, 2025 See also commentary by Júdice de Mattos Farina and Kuriki in this issue.

Purpose To present and evaluate Dafne (deep anatomical federated network), a freely available decentralized, collaborative deep learning system for the semantic segmentation of radiologic images through federated incremental learning. Materials and Methods Dafne is free software with a client-server architecture. The client side is an advanced user interface that applies the deep learning models stored on the server to the user's data and allows the user to check and refine the prediction. Incremental learning is then performed on the client's side and sent back to the server, where it is integrated into the root model. Dafne was evaluated locally by assessing the performance gain across model generations on 38 MRI datasets of the lower legs and through the analysis of real-world usage statistics (639 use cases). Results Dafne demonstrated a statistical improvement in the accuracy of semantic segmentation over time (average increase of the Dice similarity coefficient by 0.007 points per generation on the local validation set, P < .001). Qualitatively, the models showed enhanced performance on various radiologic image types, including those not present in the initial training sets, indicating good model generalizability. Conclusion Dafne showed improvement in segmentation quality over time, demonstrating potential for learning and generalization. Keywords: Segmentation, Muscular, Open Client-Server Framework Supplemental material is available for this article. © RSNA, 2025.

Purpose To develop methods for creating pseudo-contrast-enhanced US (CEUS) by using an enhanced generative adversarial network and evaluate its ability to assess tumor ablation effectiveness. Materials and Methods This retrospective study included 1030 patients who underwent thyroid nodule ablation across seven centers from January 2020 to April 2023. A generative adversarial network-based model was developed for direct pseudo-CEUS generation from B-mode US and tested on thyroid, breast, and liver ablation datasets. The reliability of pseudo-CEUS was assessed using structural similarity index (SSIM), color histogram correlation, and mean absolute percentage error against real CEUS. Additionally, a subjective evaluation system was devised to validate its clinical value. The Wilcoxon signed rank test was employed to analyze differences in the data. Results The study included 1030 patients (mean age, 46.9 years ± 12.5 [SD]; 799 female and 231 male patients). For internal test set 1, the mean SSIM was 0.89 ± 0.05, while across external test sets 1-6, mean SSIM values ranged from 0.84 ± 0.08 to 0.88 ± 0.04. Subjective assessments affirmed the method's stability and near-realistic performance in evaluating ablation effectiveness. The thyroid ablation datasets had an average identification score of 0.49 (0.5 indicates indistinguishability), while the similarity average score for all datasets was 4.75 out of 5. Radiologists' assessments of residual blood supply were nearly consistent, with no differences in defining ablation zones between real and pseudo-CEUS. Conclusion The pseudo-CEUS method demonstrated high similarity to real CEUS in evaluating tumor ablation effectiveness. Keywords: Ablation Techniques, Ultrasound, Computer Applications-Virtual Imaging Supplemental material is available for this article. © RSNA, 2025.