Radiology-Artificial Intelligence最新文献

{"title":"Human-AI Collaboration in Radiology: The Blind Spots.","authors":"Su Hwan Kim, Lisa C Adams, Benedikt Wiestler, Dennis M Hedderich","doi":"10.1148/ryai.260325","DOIUrl":"10.1148/ryai.260325","url":null,"abstract":"<p><p>Research on artificial intelligence (AI) in diagnostic radiology has focused on algorithm development and standalone performance, yet the human-AI interactions that ultimately determine the clinical value of AI are poorly understood. A narrative synthesis of human-AI interaction and the radiology AI literature highlighted three underrecognized determinants of successful human-AI collaboration in radiology. First, cognitive psychology dictates how automation bias, the framing of AI uncertainty, and AI-induced skill decay distort diagnostic reasoning. Second, the user interface (or, UI) and user experience (or, UX) of AI tools determine how the timing, salience, and documentation of AI findings shape radiologists' attention and reporting behavior; the influence of UI/UX is becoming even more critical as generative AI introduces novel applications and interaction paradigms. Third, algorithmic conformity leads radiologists to override their own judgment under medicolegal, transparency, and organizational pressures. For each domain, concrete research directions are proposed to bridge the gap between algorithmic capabilities and clinical utility. <b>Keywords:</b> Computer Applications-General, Informatics, Technology Assessment © RSNA, 2026.</p>","PeriodicalId":29787,"journal":{"name":"Radiology-Artificial Intelligence","volume":" ","pages":"e260325"},"PeriodicalIF":20.1,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148881949","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Benchmarking of AI and Radiologists for Indeterminate Lung Nodule Malignancy Risk Estimation at Screening CT: The LUNA25 Challenge.","authors":"Dré Peeters, Bogdan Obreja, Noa Antonissen, Zaigham Saghir, Ugo Pastorino, Mario Silva, Geertruida H de Bock, Hester Gietema, Fergus Gleeson, Marjolein A Heuvelmans, Stephen Lam, Geert Litjens, Firdaus Mohamed Hoesein, Cornelia Schaefer-Prokop, Ernst Scholten, Annemiek Snoeckx, Erik H F M van der Heijden, Rozemarijn Vliegenthart, Mathias Prokop, Colin Jacobs","doi":"10.1148/ryai.260179","DOIUrl":"10.1148/ryai.260179","url":null,"abstract":"<p><p>Purpose To compare the performance of an artificial intelligence (AI) system with that of radiologists for estimating malignancy risk of indeterminate-size nodules (5-15 mm) at low-dose CT (LDCT) within a standardized and transparent evaluation framework. Materials and Methods Teams participating in the AI study had access to a public dataset of 555 malignant and 5608 benign nodules on 4069 baseline LDCT scans from the National Lung Screening Trial to develop AI systems. External testing was performed on 156 malignant and 312 benign size-matched nodules, all of indeterminate size, from 463 baseline scans collected from three large European lung cancer screening trials, and the best-performing AI system (based on area under the receiver operating characteristic curve [AUC]) was selected. An observer study was conducted in which radiologists assessed 300 randomly selected nodules (100 malignant, 200 benign) from the external test set. Radiologists categorized nodules as low, intermediate, or high risk, and the threshold of intermediate or greater risk (intermediate or high-risk) was used to define a positive test result. The selected AI system was compared with radiologists on this subset using the AUC. Results The selected AI system demonstrated superior performance to the 65 radiologists' mean (AUC, 0.78 [95% CI: 0.73, 0.84] vs 0.70 [95% CI: 0.65, 0.74]; <i>P</i> = .001). With use of the intermediate risk or greater threshold, the AI system correctly classified 12% more malignant nodules at matched specificity and yielded 20% fewer false-positive results at matched sensitivity. Conclusion The selected AI system was superior to radiologists in estimating malignancy risk of indeterminate lung nodules at LDCT. <b>Keywords:</b> CT, Thorax, Lung, Observer Performance, Screening, Supervised Learning, Lung Cancer Screening, Radiologists, Artificial Intelligence, Benchmarking, Pulmonary Nodule Malignancy Risk, Deep Learning <i>Supplemental material is available for this article.</i> © RSNA, 2026 See also commentary by Júdice de Mattos Farina and Szarf in this issue.</p>","PeriodicalId":29787,"journal":{"name":"Radiology-Artificial Intelligence","volume":" ","pages":"e260179"},"PeriodicalIF":20.1,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148309725","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Deep Learning Model Incorporating Spatiotemporal Asymmetries on Longitudinal Mammograms to Predict Breast Cancer Risk.","authors":"Zhengbo Zhou, Dooman Arefan, Margarita L Zuley, Jules H Sumkin, Shandong Wu","doi":"10.1148/ryai.250888","DOIUrl":"10.1148/ryai.250888","url":null,"abstract":"<p><p>Purpose To develop and evaluate a new deep learning-based risk model that explicitly captures bilateral and longitudinal asymmetries on sequential mammograms for predicting breast cancer risk. Materials and Methods In this institutional review board-approved retrospective case-control study, the images of sequential mammographic examinations (at least two per patient, with interexamination intervals of 12-36 months)-all of which were acquired with Hologic systems-were extracted from the Cohort of Screen-Age Women - Case Control (CSAW-CC) dataset (406 patients with cancer and 6053 control participants) and an independent dataset (293 patients with cancer and 297 control participants). A novel model called STA-Risk (spatial and temporal asymmetry-based risk prediction), whose architecture incorporates side encoding, temporal encoding, and customized asymmetry loss, was constructed using these data. Fivefold cross-validation with the individual datasets and joint training with a mixed dataset were performed. Model performance was reported with the concordance index (C-index) and the time-dependent area under the receiver operating characteristic curve (AUC) (1-5 years). Results STA-Risk achieved C-indexes of 0.72 with the CSAW-CC data and 0.73 with the independent cohort data and outperformed all the compared risk models (range, 0.67-0.70 and 0.66-0.72, respectively), with consistently higher AUCs in the 1- to 5-year risk predictions. Ablation studies revealed that all three key components of STA-Risk contributed to this improved performance. Domain shifts were observed, but their effects were mitigated with joint training strategy; the resulting model achieved cross-cohort test C-indexes of 0.75 with the CSAW-CC dataset and 0.67 with the independent dataset. Conclusion The STA-Risk deep learning risk model constructed from spatiotemporal asymmetries detected on longitudinal mammograms improves breast cancer risk prediction over existing models. <b>Keywords:</b> Breast, Mammography, Technology Assessment, Experimental Investigations, Convolutional Neural Network, Diagnosis <i>Supplemental material is available for this article.</i> © RSNA, 2026.</p>","PeriodicalId":29787,"journal":{"name":"Radiology-Artificial Intelligence","volume":" ","pages":"e250888"},"PeriodicalIF":20.1,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13631677/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148621618","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Anatomy-specific Performance of CT Angiography-based AI for Anterior Circulation Occlusion: Systematic Review and Meta-Analysis.","authors":"Yike Liu, Chaofan Li, Zepeng Ren, Yuchen Hu, Chaohua Wang, Chunyan Chen, Xijin Liu","doi":"10.1148/ryai.260261","DOIUrl":"10.1148/ryai.260261","url":null,"abstract":"<p><p>Purpose To evaluate anatomy-specific diagnostic performance of CT angiography-based artificial intelligence (AI) for anterior circulation occlusion detection and negative-result implications for distal occlusions. Materials and Methods In this Preferred Reporting Items for Systematic Reviews and Meta-Analyses of Diagnostic Test Accuracy Studies (PRISMA-DTA)-compliant, prospectively registered systematic review and diagnostic meta-analysis, PubMed, Embase, and Web of Science were searched for AI-based CT angiography studies from January 1, 2019, through March 11, 2026. Pooled sensitivity and specificity were estimated for global large-vessel occlusions (LVOs), combined internal carotid artery/first segment of the middle cerebral artery (ICA/M1) occlusions, and distal second/third segments of the middle cerebral artery (M2/M3) occlusions with a bivariate random-effects model. Evidence certainty was assessed with GRADE; a supportive record-level multilevel bivariate generalized linear mixed model (GLMM) adjusted for occlusion territory, study design, algorithm type, publication year, and section thickness. Results Thirty-one reports involving 15,708 patients were included. Pooled sensitivity/specificity were 80.1%/91.9% for global LVOs, 91.6%/92.7% for ICA/M1 occlusions, and 52.7%/94.8% for M2/M3 occlusions. In the supportive adjusted model, ICA/M1 occlusions had higher sensitivity than global LVOs (OR, 2.28; <i>P</i> < .001), whereas M2/M3 occlusions had lower sensitivity (OR, 0.25; <i>P</i> < .001). Evidence certainty was moderate for ICA/M1 sensitivity and very low for M2/M3 sensitivity. For M2/M3 occlusions, a negative AI result yielded an LR- of 0.50 (95% CI, 0.28-0.73), corresponding to posttest probabilities of 17.6% and 33.3% at 30% and 50% pretest probabilities. Conclusion CT angiography-based AI performance varied by occlusion territory; lower M2/M3 sensitivity limited negative-result reliability. ©RSNA, 2026.</p>","PeriodicalId":29787,"journal":{"name":"Radiology-Artificial Intelligence","volume":" ","pages":"e260261"},"PeriodicalIF":20.1,"publicationDate":"2026-07-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148621574","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Externally Tested AI for Lung Nodule Classification: A Realistic Benchmark for an Emerging Screening Era.","authors":"Keno K Bressem, Su Hwan Kim","doi":"10.1148/ryai.260529","DOIUrl":"10.1148/ryai.260529","url":null,"abstract":"","PeriodicalId":29787,"journal":{"name":"Radiology-Artificial Intelligence","volume":"8 4","pages":"e260529"},"PeriodicalIF":20.1,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148362614","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Development and Validation of a Deep Learning-enabled Single Breath-hold Abbreviated MRI Protocol for Hepatocellular Carcinoma Diagnosis.","authors":"Yunfei Zhang, Zhijun Geng, Xianling Qian, Yuyao Xiao, Ruofan Sheng, Mengsu Zeng","doi":"10.1148/ryai.250914","DOIUrl":"10.1148/ryai.250914","url":null,"abstract":"<p><p>Purpose To develop a deep learning-enabled single breath-hold abbreviated MRI (DL-SBH-aMRI) protocol for hepatocellular carcinoma (HCC) diagnosis. Materials and Methods Patients at high risk for HCC from four institutions (January 2019-January 2025) were prospectively and retrospectively included. All patients underwent conventional complete MRI (cMRI) examinations, including precontrast T1-weighted imaging (pre-T1); postcontrast T1-weighted imaging in arterial, portal venous, and delayed phases; T2-weighted imaging (T2WI); diffusion-weighted imaging (DWI); and apparent diffusion coefficient mapping. Four generative models were trained to synthesize full MRI sequences (T2WI, DWI, apparent diffusion coefficient, arterial phase, portal venous phase, delayed phase) from pre-T1. The best-performing model was selected to generate synthetic sequences, which, combined with pre-T1 acquired from MRI, formed the DL-SBH-aMRI protocol. Image quality, perceptual realism, and lesion size measurement accuracy were evaluated for DL-SBH-aMRI versus cMRI; diagnostic performance at the patient and lesion levels was assessed using a reference standard based on histopathology and imaging findings. Results A total of 1008 patients were included (mean age ± SD, 56.8 years ± 11.8; 700 male patients). The diffusion-based generative model (Li-DiffNet) yielded the highest synthetic image quality and was selected as the backbone of DL-SBH-aMRI. DL-SBH-aMRI was noninferior to cMRI in subjective image quality scores (4.07-4.16 vs 4.18-4.19; <i>P</i> < .001). DL-SBH-aMRI demonstrated noninferior diagnostic performance (all <i>P</i> < .001) for HCC at the patient and lesion levels (sensitivity of 77.9%-88.7%; specificity of 91.6%-93.1%) compared with cMRI (sensitivity of 84.4%-92.5%; specificity of 94.1%-95.2%). Conclusion The DL-SBH-aMRI protocol may enable gadolinium-free, rapid MRI of the liver while preserving full-sequence diagnostic information for HCC diagnosis. <b>Keywords:</b> Hepatocellular Carcinoma, Deep Learning, Abbreviated MRI, Diagnostic Evaluation, Image Generation <i>Supplemental material is available for this article.</i> © The Author(s) 2026. Published by the Radiological Society of North America under a CC BY 4.0 license. See also commentary by Wang and Gu in this issue.</p>","PeriodicalId":29787,"journal":{"name":"Radiology-Artificial Intelligence","volume":" ","pages":"e250914"},"PeriodicalIF":20.1,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147975961","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"AI as a Safety Net Reader for Mammograms Classified as Normal or Benign in the French Screening Program.","authors":"Christophe Tourasse, Benoît Mesurolle, Maud Ottavy, Patricia Soler-Michel, Patrice Taourel","doi":"10.1148/ryai.250989","DOIUrl":"10.1148/ryai.250989","url":null,"abstract":"<p><p>In the French national breast cancer screening program, second reading is performed only for mammograms interpreted as negative (Breast Imaging and Reporting Data System [BI-RADS] 1-2) at first reading, representing a unique screening workflow in Europe. This retrospective study assessed whether artificial intelligence (AI) could identify a subgroup of negative screening mammograms that could safely bypass second reading. A total of 55 589 screening mammograms from 42 419 women aged 50-74 years (January 2015-December 2019) whose mammograms were initially classified as BI-RADS 1-2 were analyzed. Second reading outcomes were compared with those of a commercial AI system using a predefined binary threshold (≥5). Among these examinations, 183 of 55 589 (0.33%) were recalled at second reading, yielding 12 cancers (positive predictive value, 6.6%; cancer detection rate, 0.22 per 1000 examinations). AI classified 42 606 of 55 589 (76.6%) examinations as low risk (≤4) and 12 983 of 55 589 (23.3%) as nonlow risk (≥5). One cancer was detected in the AI-low group (one of 55 589 [0.002%]) compared with 11 in the AI-nonlow group (11 of 55 589 [0.020%]; <i>P</i> < .001). Interval cancer rates were higher in the AI-nonlow group than in the AI-low group (2.16 vs 0.47 per 1000 examinations). These findings suggest that excluding AI-low examinations from second reading could reduce workload by approximately 77% while focusing radiologist review on higher-risk cases; prospective validation is needed. <b>Keywords:</b> Mammography, Breast, Computer Aided Diagnosis (CAD) <i>Supplemental material is available for this article.</i> © RSNA, 2026 See also commentary by Yao and Chae in this issue.</p>","PeriodicalId":29787,"journal":{"name":"Radiology-Artificial Intelligence","volume":" ","pages":"e250989"},"PeriodicalIF":20.1,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148035976","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Tumor-preserving Deformable Registration of Longitudinal DCE Breast MRI during Neoadjuvant Chemotherapy.","authors":"Luyi Han, Tao Tan, Tianyu Zhang, Yuan Gao, Xin Wang, Valentina Longo, Sofía Ventura-Díaz, Anna D'Angelo, Yue Sun, Jonas Teuwen, Ritse Mann","doi":"10.1148/ryai.250789","DOIUrl":"10.1148/ryai.250789","url":null,"abstract":"<p><p>Purpose To develop a deep learning-based deformable registration method for dynamic contrast-enhanced (DCE) breast MRI that preserves tumor regions while maintaining global anatomic alignment during neoadjuvant chemotherapy (NAC) response assessment. Materials and Methods This retrospective study included internal and external cohorts of patients with breast cancer who were undergoing NAC. The internal cohort comprised patients who underwent DCE MRI from 2017 to 2020, and the external cohort was derived from the I-SPY2 trial. A conditional pyramid registration network integrating unsupervised keypoint detection with a volume-preserving mechanism was developed. Registration performance was evaluated using the Dice similarity coefficient (DSC), average landmark error, and tumor volume difference. A local-global biomarker derived from registered images was evaluated for predicting pathologic complete response (pCR) using the area under the receiver operating characteristic curve (AUC) and accuracy. Paired <i>t</i> tests were used for statistical comparisons. Results In 314 patients (all female; age, 50.6 years ± 12.0 [SD]) with 1630 scans in the internal cohort, the proposed method achieved a DSC of 0.95 ± 0.02, an average landmark error of 5.35 mm ± 3.46, and a tumor volume difference of 11.0% ± 10.7. In 100 patients (all female; age, 48.5 years ± 12.3) with 372 scans in the external cohort, the method achieved a DSC of 0.91 ± 0.09 and a tumor volume difference of 15.5% ± 13.8. Improvements in landmark distance and tumor preservation were statistically significant (<i>P</i> < .05) compared with most methods. For pCR prediction, incorporation of the proposed biomarker achieved an AUC of 0.81 ± 0.04 and an accuracy of 72.1% ± 5.0. Conclusion The proposed framework improved anatomic alignment while preserving tumor volume in longitudinal DCE breast MRI during NAC response assessment and enabled a registration-based biomarker for predicting treatment response. <b>Keywords:</b> MR Imaging, Image Postprocessing, Breast, Neural Networks, Radiomics, Prognosis, DCE Breast MRI, Deep Learning, Deformable Registration, Neoadjuvant Chemotherapy, Unsupervised Keypoint Detection <i>Supplemental material is available for this article.</i> © RSNA, 2026 See also the commentary by Zhang in this issue.</p>","PeriodicalId":29787,"journal":{"name":"Radiology-Artificial Intelligence","volume":" ","pages":"e250789"},"PeriodicalIF":20.1,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147975898","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"When One Sequence Is Enough-And When It Isn't.","authors":"Kang Wang, Hanxue Gu","doi":"10.1148/ryai.260501","DOIUrl":"10.1148/ryai.260501","url":null,"abstract":"","PeriodicalId":29787,"journal":{"name":"Radiology-Artificial Intelligence","volume":"8 4","pages":"e260501"},"PeriodicalIF":20.1,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148309686","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Impact of Exposure Parameters on Deep Learning Models in Chest Radiography and Implications for Deployment.","authors":"Han-Jay Shu, Shun-Ting Chang, Rodrigo Rosa Gameiro, Judy Wawira Gichoya, Leo Anthony Celi, Po-Chih Kuo","doi":"10.1148/ryai.250731","DOIUrl":"10.1148/ryai.250731","url":null,"abstract":"<p><p>Purpose To investigate whether deep learning models trained on chest radiographs rely on radiographic exposure parameters as shortcut features and to quantify the resulting biases under controlled confounding and natural exposure regimes. Materials and Methods In this retrospective study, chest radiographs from MIMIC-CXR (January 2011-December 2016), the Medical Imaging and Data Resource Center (August 2020-May 2022), and EmoryCXR (September 2008-February 2023) were analyzed for pneumothorax detection, COVID-19 diagnosis, and race classification. Dataset-provided labels served as the reference standard. Three exposure parameters (ExposureTime, XRayTubeCurrent, and ExposureInuAs) were extracted from Digital Imaging and Communications in Medicine metadata. Models were trained under biased and balanced exposure label alignments and evaluated on matched and reversed distributions. A priori screening additionally identified high-risk exposure regimes. The area under the receiver operating characteristic curve (AUC) was compared using the DeLong test. Results A total of 727 604 chest radiographs in 240 681 patients (mean age ± SD, 60 years ± 17; 126 432 men, 114 128 women) were analyzed. For pneumothorax detection, AUC decreased from 0.94 (95% CI: 0.94, 0.95) to 0.56 (95% CI: 0.55, 0.58) on mismatched exposure distributions (ΔAUC, -0.38; <i>P</i> < .001). Similar declines were observed for COVID-19 (ΔAUC, -0.33; <i>P</i> < .001) and race classification (ΔAUC, -0.09; <i>P</i> < .001). The a priori exposure-regimen screening revealed high-risk regimes within the natural distribution that were associated with reduced model performance compared with typical exposures. Conclusion Deep learning models trained on chest radiographs may exploit exposure parameters as shortcut features; exposure-regimen audits may flag high-risk conditions before clinical deployment. <b>Keywords:</b> Computer Aided Diagnosis (CAD), Lung, Feature Detection, Diagnosis, Convolutional Neural Network (CNN), Chest Radiograph, Exposure Parameters, Deep Learning, Fairness, Shortcut Learning <i>Supplemental material is available for this article.</i> © RSNA, 2026 See also commentary by Le Guellec and Chassagnon in this issue.</p>","PeriodicalId":29787,"journal":{"name":"Radiology-Artificial Intelligence","volume":" ","pages":"e250731"},"PeriodicalIF":20.1,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13430028/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148309648","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}