Academic Radiology最新文献

{"title":"Integrating CT Radiomics and Machine Learning for Preoperative Assessment of STAS Grading and Prognostic Stratification in Lung Adenocarcinoma","authors":"Qiuxing Chen MS, Siqi Qu MS, Fang Chen MS, Shulin Li MS, Zhihe Yang MS, Qijia Han MB, Zhu Ai MS, Peng Wang MD, Kun Ma PhD, Zhiming Xiang MD","doi":"10.1016/j.acra.2026.06.057","DOIUrl":"10.1016/j.acra.2026.06.057","url":null,"abstract":"<div><h3>Rationale and Objectives</h3><div>The distance of spread through air spaces (STAS) dissemination is associated with prognosis in lung adenocarcinoma (LUAD). This study aimed to develop a computed tomography (CT) radiomics-based machine learning model to predict STAS grade.</div></div><div>The distance of spread through air spaces (STAS) dissemination is associated with prognosis in lung adenocarcinoma (LUAD). This study aimed to develop a computed tomography (CT) radiomics-based machine learning model to predict STAS grade.</div><div><h3>Materials and Methods</h3><div>Retrospective data included 293 LUAD patients for training/internal validation and 92 from the National Lung Screening Trial for external validation. STAS, defined as tumor cells beyond the tumor edge, was stratified by a 2500 µm threshold into grade I and II. Cox regression assessed associations between STAS grade and recurrence-free survival (RFS). Radiomics features were selected by minimum redundancy maximum relevance (MRMR), and 11 machine learning classifiers predicted STAS presence and grade. Independent clinic-radiological predictors were identified by univariable and multivariable analyses, and logistic regression combined these with the best radiomics model to build a combined model. Performance was evaluated by area under the curve (AUC) and accuracy (ACC).</div></div><div>Retrospective data included 293 LUAD patients for training/internal validation and 92 from the National Lung Screening Trial for external validation. STAS, defined as tumor cells beyond the tumor edge, was stratified by a 2500 µm threshold into grade I and II. Cox regression assessed associations between STAS grade and recurrence-free survival (RFS). Radiomics features were selected by minimum redundancy maximum relevance (MRMR), and 11 machine learning classifiers predicted STAS presence and grade. Independent clinic-radiological predictors were identified by univariable and multivariable analyses, and logistic regression combined these with the best radiomics model to build a combined model. Performance was evaluated by area under the curve (AUC) and accuracy (ACC).</div><div><h3>Results</h3><div>STAS grade II was independently associated with poorer RFS (<em>P =</em> 0.037), while grade I was not. For STAS detection, the random forest model achieved AUCs of 0.888, 0.886 and 0.779 with accuracies of 81.5%, 85.2%, and 69.6%; the combined model performed similarly (AUCs 0.877, 0.894, 0.802; accuracies 77.6%, 85.2%, 72.8%). For STAS grading, logistic regression yielded AUCs of 0.845, 0.852, and 0.798 (accuracies 78.2%, 70.4%, 77.1%), with the combined model comparable (AUCs 0.845, 0.855, 0.796; accuracies 78.2%, 70.4%, 77.1%).</div></div><div>STAS grade II was independently associated with poorer RFS (<em>P =</em> 0.037), while grade I was not. For STAS detection, the random forest model achieved AUCs of 0.888, 0.886 and 0.779 with accuracies of 81.5%, 85.2%, and 69.6%; the combined model perfor","PeriodicalId":50928,"journal":{"name":"Academic Radiology","volume":"33 10","pages":"Pages 4634-4646"},"PeriodicalIF":4.7,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148474065","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Implementation of an Early Exposure Program for College Students as a Strategy to Increase the Medical Imaging Workforce.","authors":"Felipe Godinez, Ramsey D Badawi, Cyrus Bateni, Sarah McKenney, Shannon M Navarro, Elizabeth Morris, Tracy Yarbrough","doi":"10.1016/j.acra.2026.09.008","DOIUrl":"https://doi.org/10.1016/j.acra.2026.09.008","url":null,"abstract":"<p><strong>Rationale and objectives: </strong>Radiology careers require lengthy academic preparation, yet many students remain unaware of them until well into their studies, limiting early trajectory-setting and sustained motivation. We describe the architecture and implementation of the Radiology and Imaging Science Scholars program (RaISS), a pipeline program developed by the Department of Radiology at UC Davis Health to expose community college and early undergraduate students to radiology careers.</p><p><strong>Materials and methods: </strong>RaISS treats community college and early undergraduate students as junior radiology colleagues, granting departmental badges and internal system access over a structured, six-month experience combining clinical shadowing, research engagement, mentorship, and professional development. Participants were recruited regionally and selected for interest in healthcare or biomedical sciences and limited prior exposure to academic medicine. Impact was assessed through post-program surveys of radiology understanding, career-pathway awareness, and confidence in further education.</p><p><strong>Results: </strong>Across the first two cohorts (2024-2025), applications rose from 6 to 9 (31 received for 2026). Survey response rate was 58% (7 of 12); all respondents strongly agreed that the program increased their awareness of radiology/imaging careers, and 100% reported it influenced career consideration. One 2024 participant has since matriculated to medical school. Mentor survey completion was 67% (8 of 12); all 2024 mentors expressed interest in returning. Onboarding took about 4 weeks and was a key barrier.</p><p><strong>Conclusion: </strong>Structured early exposure to radiology through mentorship, clinical observation, and research engagement may broaden awareness of imaging careers among students with limited access to academic medicine, supporting efforts to strengthen the radiology workforce pipeline.</p>","PeriodicalId":50928,"journal":{"name":"Academic Radiology","volume":" ","pages":""},"PeriodicalIF":4.7,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149117830","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Promotion from Associate Professor to Full Professor Should Not Be Monolithic: A National Bibliometric Study by Radiology Subspecialty","authors":"Dhairya A. Lakhani M.D.,&nbsp;Mahla Radmard MD,&nbsp;David M. Yousem MD, MBA","doi":"10.1016/j.acra.2026.06.004","DOIUrl":"10.1016/j.acra.2026.06.004","url":null,"abstract":"&lt;div&gt;&lt;h3&gt;Background&lt;/h3&gt;&lt;div&gt;Academic promotion from associate professor (AP) to full professor (FP) in radiology relies heavily on bibliometrics, yet no subspecialty-specific benchmarks exist to guide promotion committees or faculty career planning. Applying uniform productivity standards across diverse subspecialties may disadvantage faculty in certain fields and fail to account for legitimate variations in research output. Hence, we aim to establish subspecialty-specific bibliometric benchmarks for promotion readiness from AP to FP in academic radiology and quantify the promotion gap across subspecialties.&lt;/div&gt;&lt;/div&gt;&lt;div&gt;Academic promotion from associate professor (AP) to full professor (FP) in radiology relies heavily on bibliometrics, yet no subspecialty-specific benchmarks exist to guide promotion committees or faculty career planning. Applying uniform productivity standards across diverse subspecialties may disadvantage faculty in certain fields and fail to account for legitimate variations in research output. Hence, we aim to establish subspecialty-specific bibliometric benchmarks for promotion readiness from AP to FP in academic radiology and quantify the promotion gap across subspecialties.&lt;/div&gt;&lt;div&gt;&lt;h3&gt;Methods&lt;/h3&gt;&lt;div&gt;We analyzed 1931 APs and 1759 FPs across nine radiology subspecialties at 136 US institutions between September 2025 and January 2026. Bibliometric data (publications, citations, h-index, first/last author publications) were obtained from Scopus. Upper-tier AP percentiles (60th-90th) defined \"promotion readiness\" benchmarks, while lower-tier FP percentiles (10th-40th) represented \"entry thresholds.\" The promotion gap was quantified by comparing percentile distributions and calculating median bibliometric differences between ranks.&lt;/div&gt;&lt;/div&gt;&lt;div&gt;We analyzed 1931 APs and 1759 FPs across nine radiology subspecialties at 136 US institutions between September 2025 and January 2026. Bibliometric data (publications, citations, h-index, first/last author publications) were obtained from Scopus. Upper-tier AP percentiles (60th-90th) defined \"promotion readiness\" benchmarks, while lower-tier FP percentiles (10th-40th) represented \"entry thresholds.\" The promotion gap was quantified by comparing percentile distributions and calculating median bibliometric differences between ranks.&lt;/div&gt;&lt;div&gt;&lt;h3&gt;Results&lt;/h3&gt;&lt;div&gt;Promotion readiness at the 80th percentile varied 4.2-fold for publications [154 (Neurointerventional) vs. 37 (Breast)], 4.5-fold for citations [3598 (Nuclear Medicine) vs. 799 (Musculoskeletal)], and 2.2-fold for h-index [31 (Neurointerventional) vs. 14 (Breast)] across subspecialties. An AP at the 80th percentile in Pediatric Radiology would rank at the 49th percentile among FPs (smallest gap), while one in Musculoskeletal Radiology would rank at the 29th percentile (largest gap)—a 21-percentile-point difference. The AP 80th percentile exceeded the FP 20th percentile for every subspecialty. Achieving median FP bibliom","PeriodicalId":50928,"journal":{"name":"Academic Radiology","volume":"33 10","pages":"Pages 4207-4217"},"PeriodicalIF":4.7,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148341012","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"The 2025 National Institute of Health Budget Cuts and the Future of Radiology Innovation: Implications for the Physician-Scientist Pipeline and Research Paradigms","authors":"Ali Rashidi MD,&nbsp;Patricia Balthazar MD, MPH,&nbsp;Elizabeth A. Krupinski PhD, FSPIE, FSIIM, FATA, FAIMBE, FAUR","doi":"10.1016/j.acra.2026.07.009","DOIUrl":"10.1016/j.acra.2026.07.009","url":null,"abstract":"<div><div>The United States National Institutes of Health (NIH) serves as the cornerstone of biomedical research funding. Yet, radiology departments receive a disproportionately small share of this support, concentrated among senior investigators at large academic institutions. Beginning in January 2025, the current administration proposed an approximately 40% reduction in federal research funding, equivalent to a $20 billion decrease, and implemented the termination or freezing of over 2500 grants, representing $1.4 billion in research loss, alongside a proposed 15% mandatory cap on indirect cost recovery. This perspective examines the potential consequences of these administrative changes on the radiology physician-scientist pipeline and the future of imaging research and innovation. Research Project Grant success rates are projected to decline from 18.5% in fiscal year (FY) 2024 to a historic low of 7.3% by FY 2026, compounding the already widening gap between established and early-career investigators. Radiology-specific training pathways remain critically undersubscribed, with the Holman Research Pathway averaging only 1.6 graduates per year over 20 years. The systematic erosion of NIH training grants and dropping funding success rates risk permanent structural collapse of the radiology physician-scientist workforce, and jeopardizing United States leadership in imaging science, artificial intelligence, and translational innovation.</div></div><div>The United States National Institutes of Health (NIH) serves as the cornerstone of biomedical research funding. Yet, radiology departments receive a disproportionately small share of this support, concentrated among senior investigators at large academic institutions. Beginning in January 2025, the current administration proposed an approximately 40% reduction in federal research funding, equivalent to a $20 billion decrease, and implemented the termination or freezing of over 2500 grants, representing $1.4 billion in research loss, alongside a proposed 15% mandatory cap on indirect cost recovery. This perspective examines the potential consequences of these administrative changes on the radiology physician-scientist pipeline and the future of imaging research and innovation. Research Project Grant success rates are projected to decline from 18.5% in fiscal year (FY) 2024 to a historic low of 7.3% by FY 2026, compounding the already widening gap between established and early-career investigators. Radiology-specific training pathways remain critically undersubscribed, with the Holman Research Pathway averaging only 1.6 graduates per year over 20 years. The systematic erosion of NIH training grants and dropping funding success rates risk permanent structural collapse of the radiology physician-scientist workforce, and jeopardizing United States leadership in imaging science, artificial intelligence, and translational innovation.</div>","PeriodicalId":50928,"journal":{"name":"Academic Radiology","volume":"33 10","pages":"Pages 4198-4203"},"PeriodicalIF":4.7,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148609729","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Multimodal Ultrasound Radiomics Model Integrating Contrast-Enhanced Phases: Retrospective Temporal Test Cohort Evaluation and Comparison with Sonographers for Cervical Lymph Node Characterization","authors":"Mingru Gao,&nbsp;Qian Lv,&nbsp;Haohao Hao,&nbsp;Ning Zhao,&nbsp;Yuping Guo,&nbsp;Kuo Miao","doi":"10.1016/j.acra.2026.06.048","DOIUrl":"10.1016/j.acra.2026.06.048","url":null,"abstract":"&lt;div&gt;&lt;h3&gt;Rationale and Objectives&lt;/h3&gt;&lt;div&gt;To develop a multimodal ultrasound radiomics model based on conventional ultrasound and three-phase contrast-enhanced ultrasound (CEUS) for distinguishing benign from malignant cervical lymph nodes, and to evaluate its performance in a patient-level temporal test cohort against sonographers with different experience levels.&lt;/div&gt;&lt;/div&gt;&lt;div&gt;To develop a multimodal ultrasound radiomics model based on conventional ultrasound and three-phase contrast-enhanced ultrasound (CEUS) for distinguishing benign from malignant cervical lymph nodes, and to evaluate its performance in a patient-level temporal test cohort against sonographers with different experience levels.&lt;/div&gt;&lt;div&gt;&lt;h3&gt;Materials and Methods&lt;/h3&gt;&lt;div&gt;This retrospective study included 245 patients in the development cohort (August 2022–December 2024) and 60 subsequent patients in the temporal test cohort (January–December 2025). One target lymph node was analyzed per patient, and the two cohorts were strictly non-overlapping at the patient level. For each case, matched two-dimensional ultrasound (2D-US), color Doppler flow imaging (CDFI), and arterial-, parenchymal-, and clearance-phase CEUS key frames were collected. Deep features were extracted from manually cropped ROIs using a frozen pre-trained ResNet50. Feature selection, hyperparameter tuning, and decision-threshold locking were completed only within the development cohort; the temporal test cohort was used once for final reporting. Sonographers reviewed the same static key-frame image set in randomized order under blinded conditions.&lt;/div&gt;&lt;/div&gt;&lt;div&gt;This retrospective study included 245 patients in the development cohort (August 2022–December 2024) and 60 subsequent patients in the temporal test cohort (January–December 2025). One target lymph node was analyzed per patient, and the two cohorts were strictly non-overlapping at the patient level. For each case, matched two-dimensional ultrasound (2D-US), color Doppler flow imaging (CDFI), and arterial-, parenchymal-, and clearance-phase CEUS key frames were collected. Deep features were extracted from manually cropped ROIs using a frozen pre-trained ResNet50. Feature selection, hyperparameter tuning, and decision-threshold locking were completed only within the development cohort; the temporal test cohort was used once for final reporting. Sonographers reviewed the same static key-frame image set in randomized order under blinded conditions.&lt;/div&gt;&lt;div&gt;&lt;h3&gt;Results&lt;/h3&gt;&lt;div&gt;The CEUS-integrated SVM model (Model 2-SVM) showed the best overall performance among the tested models. In the temporal test cohort, it achieved an area under the curve (AUC) of 0.851 (95% CI: 0.762–0.940), sensitivity of 84.8% (95% CI: 69.1–93.3), specificity of 66.7% (95% CI: 47.8–81.4), accuracy of 76.7% (95% CI: 64.6–85.6), PPV of 75.7% (95% CI: 59.9–86.6), and NPV of 78.3% (95% CI: 58.1–90.3). The senior and junior sonographers achieved AUCs of 0.826 and 0.783, respectivel","PeriodicalId":50928,"journal":{"name":"Academic Radiology","volume":"33 10","pages":"Pages 4583-4592"},"PeriodicalIF":4.7,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148671300","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Injuries to the Craniocervical Junction: A 3D Approach to Resident Education for Radiological Imaging","authors":"Talia Mastalski MA,&nbsp;Elana B. Smith MD,&nbsp;Andrew Sinensky MD,&nbsp;Jeff Day MA, MD","doi":"10.1016/j.acra.2026.06.050","DOIUrl":"10.1016/j.acra.2026.06.050","url":null,"abstract":"&lt;div&gt;&lt;h3&gt;Rationale and Objectives&lt;/h3&gt;&lt;div&gt;Injuries to the craniocervical junction (CCJ) are notoriously challenging to diagnose on imaging, often misdiagnosed or overlooked, leading to increased morbidity and mortality. Difficulties in visualizing anatomy and a lack of learning materials illustrate the need for clear, accessible resources for residents. This study’s purpose was to create novel visual training resources and assess the effectiveness of still images and animated multimedia resources to determine which was most effective in helping residents learn CCJ injuries and which was more engaging.&lt;/div&gt;&lt;/div&gt;&lt;div&gt;Injuries to the craniocervical junction (CCJ) are notoriously challenging to diagnose on imaging, often misdiagnosed or overlooked, leading to increased morbidity and mortality. Difficulties in visualizing anatomy and a lack of learning materials illustrate the need for clear, accessible resources for residents. This study’s purpose was to create novel visual training resources and assess the effectiveness of still images and animated multimedia resources to determine which was most effective in helping residents learn CCJ injuries and which was more engaging.&lt;/div&gt;&lt;div&gt;&lt;h3&gt;Materials and Methods&lt;/h3&gt;&lt;div&gt;We created still images and an animation of a CCJ injury that related a 3D model to computed tomography/MR imaging. We tested its efficacy using a Qualtrics survey distributed to medical students and trainees involved in the treatment and diagnosis of CCJ injuries. Participants were randomized into two groups– still images and still images + animation. Participants completed a pre-test, viewed a module, and completed an engagement survey, post-test, and comparison survey.&lt;/div&gt;&lt;/div&gt;&lt;div&gt;We created still images and an animation of a CCJ injury that related a 3D model to computed tomography/MR imaging. We tested its efficacy using a Qualtrics survey distributed to medical students and trainees involved in the treatment and diagnosis of CCJ injuries. Participants were randomized into two groups– still images and still images + animation. Participants completed a pre-test, viewed a module, and completed an engagement survey, post-test, and comparison survey.&lt;/div&gt;&lt;div&gt;&lt;h3&gt;Results&lt;/h3&gt;&lt;div&gt;We tested 56 participants. There was significant improvement (p = 0.000064) between pre- and post-test scores for both still images and still images + animation groups. Animation did not significantly affect learning (p = 0.52) or engagement (p = 0.18), suggesting animation neither helped nor harmed learning or engagement. When participants were asked if they prefer learning from still images or still images + animation, still images + animation was favored by 66.07% for knowledge retention and 69.64% for engagement.&lt;/div&gt;&lt;/div&gt;&lt;div&gt;We tested 56 participants. There was significant improvement (p = 0.000064) between pre- and post-test scores for both still images and still images + animation groups. Animation did not significantly affect learning (p = ","PeriodicalId":50928,"journal":{"name":"Academic Radiology","volume":"33 10","pages":"Pages 4145-4159"},"PeriodicalIF":4.7,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148537697","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Foundation Model-empowered Multimodal Prediction of Platinum Resistance in High-grade Serous Ovarian Cancer: A Multicenter Retrospective Study","authors":"Qiu Bi MD,&nbsp;Linhao Qu PHD,&nbsp;Yang Liu MD,&nbsp;Qinqing Wang MM,&nbsp;Qingyin Meng MM,&nbsp;Jing Yang MD,&nbsp;Yunzhu Wu PHD,&nbsp;Conghui Ai MD,&nbsp;Kun Miao MM,&nbsp;Li Wu MD","doi":"10.1016/j.acra.2026.06.024","DOIUrl":"10.1016/j.acra.2026.06.024","url":null,"abstract":"&lt;div&gt;&lt;h3&gt;Rationale and Objectives&lt;/h3&gt;&lt;div&gt;To develop and validate a foundation model (FM)-empowered multimodal framework to predict platinum resistance in patients with high-grade serous ovarian cancer (HGSOC).&lt;/div&gt;&lt;/div&gt;&lt;div&gt;To develop and validate a foundation model (FM)-empowered multimodal framework to predict platinum resistance in patients with high-grade serous ovarian cancer (HGSOC).&lt;/div&gt;&lt;div&gt;&lt;h3&gt;Materials and Methods&lt;/h3&gt;&lt;div&gt;This multicenter retrospective cohort study included 744 HGSOC patients who underwent primary debulking surgery across four centers. Perioperative multimodal data were collected, including clinical variables, preoperative pelvic magnetic resonance imaging (MRI), and postoperative hematoxylin and eosin (H&amp;E)-stained whole slide images (WSIs). Pretrained radiology and pathology FMs were used as frozen feature extractors, and attention-based aggregation was applied to summarize variable-length MRI slices and WSI tiles. A cross-modal fusion module integrated modalities and supported missing-modality inference. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) and compared with late-fusion (LF) baselines.&lt;/div&gt;&lt;/div&gt;&lt;div&gt;This multicenter retrospective cohort study included 744 HGSOC patients who underwent primary debulking surgery across four centers. Perioperative multimodal data were collected, including clinical variables, preoperative pelvic magnetic resonance imaging (MRI), and postoperative hematoxylin and eosin (H&amp;E)-stained whole slide images (WSIs). Pretrained radiology and pathology FMs were used as frozen feature extractors, and attention-based aggregation was applied to summarize variable-length MRI slices and WSI tiles. A cross-modal fusion module integrated modalities and supported missing-modality inference. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC) and compared with late-fusion (LF) baselines.&lt;/div&gt;&lt;div&gt;&lt;h3&gt;Results&lt;/h3&gt;&lt;div&gt;In the internal validation cohort, unimodal and multimodal FM variants achieved AUCs ranging from 0.708 to 0.780, and in external test cohorts A, B, and C achieved AUCs ranging from 0.653 to 0.790. The trimodal FM (clinical+MRI+WSI) achieved the highest AUC in all cohorts (internal validation: 0.780; external cohorts A, B, and C: 0.779, 0.781, and 0.790, respectively), outperforming the trimodal LF model (internal validation: 0.753; external cohorts A, B, and C: 0.729, 0.715, and 0.730, respectively).&lt;/div&gt;&lt;/div&gt;&lt;div&gt;In the internal validation cohort, unimodal and multimodal FM variants achieved AUCs ranging from 0.708 to 0.780, and in external test cohorts A, B, and C achieved AUCs ranging from 0.653 to 0.790. The trimodal FM (clinical+MRI+WSI) achieved the highest AUC in all cohorts (internal validation: 0.780; external cohorts A, B, and C: 0.779, 0.781, and 0.790, respectively), outperforming the trimodal LF model (internal validation: 0.753; external cohorts ","PeriodicalId":50928,"journal":{"name":"Academic Radiology","volume":"33 10","pages":"Pages 4479-4488"},"PeriodicalIF":4.7,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148474118","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Artificial Intelligence–Based Computer-Aided Detection in Breast Cancer Diagnosis: Variation by Breast Density, Imaging Feature, and Tumor Characteristics","authors":"Natalia Eugene MD,&nbsp;Chirag Parghi MD, MBA,&nbsp;Jafer Elabeid MBBS,&nbsp;Jennifer Pantleo RN,&nbsp;Nelda Gonzalez MS,&nbsp;Avishkar Sharma MD,&nbsp;Zi Zhang MD, MSCE","doi":"10.1016/j.acra.2026.06.020","DOIUrl":"10.1016/j.acra.2026.06.020","url":null,"abstract":"&lt;div&gt;&lt;h3&gt;Rationale and Objectives&lt;/h3&gt;&lt;div&gt;To determine whether an FDA-approved artificial intelligence computer-aided detection and diagnosis (AI-CAD) system assigns varying case scores based on imaging features, tumor characteristics, and breast density.&lt;/div&gt;&lt;/div&gt;&lt;div&gt;To determine whether an FDA-approved artificial intelligence computer-aided detection and diagnosis (AI-CAD) system assigns varying case scores based on imaging features, tumor characteristics, and breast density.&lt;/div&gt;&lt;div&gt;&lt;h3&gt;Materials and Methods&lt;/h3&gt;&lt;div&gt;This retrospective multisite study included patients undergoing biopsy after abnormal screening tomosynthesis across four U.S. states in 2021. A commercially available AI-CAD tool generated case-level scores (range 0–100). Imaging features were classified as calcifications, mass, mass with calcifications, architectural distortion/asymmetry, or nodal-only metastasis. Breast density was dichotomized as dense or non-dense. Tumor size, grade and pathology were also collected. Associations were assessed using univariable and multivariable linear regression.&lt;/div&gt;&lt;/div&gt;&lt;div&gt;This retrospective multisite study included patients undergoing biopsy after abnormal screening tomosynthesis across four U.S. states in 2021. A commercially available AI-CAD tool generated case-level scores (range 0–100). Imaging features were classified as calcifications, mass, mass with calcifications, architectural distortion/asymmetry, or nodal-only metastasis. Breast density was dichotomized as dense or non-dense. Tumor size, grade and pathology were also collected. Associations were assessed using univariable and multivariable linear regression.&lt;/div&gt;&lt;div&gt;&lt;h3&gt;Results&lt;/h3&gt;&lt;div&gt;Breast cancer was diagnosed in 735/3899 patients. The mean age and case score for patients with breast cancer were 65.3±11.7 years and 78.1±22.9, respectively, both significantly higher than those of non-malignant cases (58.9±11.1 years and 31.8±25.9; &lt;em&gt;p&lt;/em&gt;&lt;0.001). Scores were highest for calcifications (91.6±15.2) and mass with calcifications (90.5±16.3), both significantly higher than mass alone (79.1±23.6; &lt;em&gt;p&lt;/em&gt;&lt;0.01). In univariable analysis, ductal carcinoma in-situ (DCIS) had higher scores than invasive cancers (&lt;em&gt;p&lt;/em&gt;&lt;0.05). However, in multivariable analysis, higher scores were associated with non-dense breasts, higher tumor grade, larger tumor size, older age, and specific imaging features, such as calcifications, whereas the association with pathologic subtype was attenuated.&lt;/div&gt;&lt;/div&gt;&lt;div&gt;Breast cancer was diagnosed in 735/3899 patients. The mean age and case score for patients with breast cancer were 65.3±11.7 years and 78.1±22.9, respectively, both significantly higher than those of non-malignant cases (58.9±11.1 years and 31.8±25.9; &lt;em&gt;p&lt;/em&gt;&lt;0.001). Scores were highest for calcifications (91.6±15.2) and mass with calcifications (90.5±16.3), both significantly higher than mass alone (79.1±23.6; &lt;em&gt;p&lt;/em&gt;&lt;0.01). In univariable analysis, ","PeriodicalId":50928,"journal":{"name":"Academic Radiology","volume":"33 10","pages":"Pages 4349-4360"},"PeriodicalIF":4.7,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148450487","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Automated MRI Liver Segmentation For Accurate Quantification Of Hepatic Steatosis","authors":"Xiaodie Wei MB,&nbsp;Shi Qi MS,&nbsp;Sitong Chen MS,&nbsp;Lixia Qiu MD,&nbsp;Xinxin Wang MD,&nbsp;Jing Zhao MD,&nbsp;Jingjing Cui MS,&nbsp;Li Cao MB,&nbsp;Jing Zhang MD","doi":"10.1016/j.acra.2026.06.036","DOIUrl":"10.1016/j.acra.2026.06.036","url":null,"abstract":"&lt;div&gt;&lt;h3&gt;Rationale and Objectives&lt;/h3&gt;&lt;div&gt;MRI-proton density fat fraction (MRI-PDFF) is widely applied in clinical practice for hepatic fat quantification. However, the conventional manual region of interest (ROI) method is time-consuming and operator-dependent, and the diagnostic accuracy of artificial intelligence (AI)-based models for hepatic steatosis assessment remains to be fully clarified. This study aimed to evaluate the diagnostic performance of an AI-based whole liver segmentation (WLS) model for histological hepatic steatosis grading, and to technically validate its segmentation performance and agreement with manual ROI-based PDFF measurement.&lt;/div&gt;&lt;/div&gt;&lt;div&gt;MRI-proton density fat fraction (MRI-PDFF) is widely applied in clinical practice for hepatic fat quantification. However, the conventional manual region of interest (ROI) method is time-consuming and operator-dependent, and the diagnostic accuracy of artificial intelligence (AI)-based models for hepatic steatosis assessment remains to be fully clarified. This study aimed to evaluate the diagnostic performance of an AI-based whole liver segmentation (WLS) model for histological hepatic steatosis grading, and to technically validate its segmentation performance and agreement with manual ROI-based PDFF measurement.&lt;/div&gt;&lt;div&gt;&lt;h3&gt;Materials and Methods&lt;/h3&gt;&lt;div&gt;A total of 538 adults who underwent MRI-PDFF examinations were enrolled. A VBB-Net segmentation model was developed in a training cohort of 372 patients (418 examinations). Validation was performed in a consecutive cohort of 166 adults who underwent liver biopsy for suspected metabolic dysfunction-associated steatotic liver disease (MASLD). Histological steatosis grading (S0–S3) was used as the reference standard.&lt;/div&gt;&lt;/div&gt;&lt;div&gt;A total of 538 adults who underwent MRI-PDFF examinations were enrolled. A VBB-Net segmentation model was developed in a training cohort of 372 patients (418 examinations). Validation was performed in a consecutive cohort of 166 adults who underwent liver biopsy for suspected metabolic dysfunction-associated steatotic liver disease (MASLD). Histological steatosis grading (S0–S3) was used as the reference standard.&lt;/div&gt;&lt;div&gt;&lt;h3&gt;Results&lt;/h3&gt;&lt;div&gt;The mean Dice coefficients were 0.94 ± 0.05 in the training cohort and 0.93 ± 0.04 in the validation cohort. Margins of error were 0.794% for AI-WLS-PDFF and 0.821% for ROI-PDFF. The AUROCs (95% CI) of AI-WLS-PDFF for diagnosing S0-S3 were 0.995 (95% CI: 0.989, 1.000), 0.951 (95% CI: 0.921, 0.980), and 0.928 (95% CI: 0.888, 0.968), respectively. AI-WLS-PDFF showed excellent agreement with ROI-PDFF (ICC, 0.996; 95% CI: 0.995, 0.997) with minimal bias (mean difference, -0.06%; 95% agreement limits: -1.59% to 1.47%).&lt;/div&gt;&lt;/div&gt;&lt;div&gt;The mean Dice coefficients were 0.94 ± 0.05 in the training cohort and 0.93 ± 0.04 in the validation cohort. Margins of error were 0.794% for AI-WLS-PDFF and 0.821% for ROI-PDFF. The AUROCs (95% CI) of AI-WLS-PDFF for diagnosing S0-S3 ","PeriodicalId":50928,"journal":{"name":"Academic Radiology","volume":"33 10","pages":"Pages 4257-4264"},"PeriodicalIF":4.7,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148457759","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Differentiation of HER2-Zero, -Low, and -Overexpressing Breast Cancer Using Amide Proton Transfer-Weighted Magnetic Resonance Imaging: A Single-center Preliminary Study","authors":"Ting Zhan PhD,&nbsp;Yinping Leng PhD,&nbsp;XinYi Liu BM,&nbsp;Jiankun Dai PhD,&nbsp;Lianggeng Gong PhD","doi":"10.1016/j.acra.2026.05.015","DOIUrl":"10.1016/j.acra.2026.05.015","url":null,"abstract":"&lt;div&gt;&lt;ul&gt;&lt;li&gt;&lt;span&gt;•&lt;/span&gt;&lt;span&gt;&lt;div&gt;Breast cancers exhibiting distinct HER2 expression levels demonstrate unique clinical and MRI morphological features, with significant differences also observed in quantitative parameters derived from DWI and APTWI.&lt;/div&gt;&lt;/span&gt;&lt;/li&gt;&lt;li&gt;&lt;span&gt;•&lt;/span&gt;&lt;span&gt;&lt;div&gt;MTRasym_3.5ppm demonstrates exceptional diagnostic performance for HER2 subtype classification in invasive breast cancer, serving as a key component within an imaging-derived framework to refine HER2 stratification and optimize personalized treatment strategies.&lt;/div&gt;&lt;/span&gt;&lt;/li&gt;&lt;li&gt;&lt;span&gt;•&lt;/span&gt;&lt;span&gt;&lt;div&gt;This comprehensive model, combining MRI qualitative and quantitative features, enables reliable tripartite HER2 stratification, providing a non-invasive, non-contrast alternative for patients with uncertain IHC/FISH results or contraindications to contrast-enhanced MRI, thus reducing the need for repeated biopsies.&lt;/div&gt;&lt;/span&gt;&lt;/li&gt;&lt;/ul&gt;&lt;/div&gt;&lt;div&gt;&lt;h3&gt;Rationale and Objectives&lt;/h3&gt;&lt;div&gt;To assess amide proton transfer-weighted imaging (APTWI) for identifying HER2 expression in breast cancer (BC) and to evaluate multiparametric model performance.&lt;/div&gt;&lt;/div&gt;&lt;div&gt;To assess amide proton transfer-weighted imaging (APTWI) for identifying HER2 expression in breast cancer (BC) and to evaluate multiparametric model performance.&lt;/div&gt;&lt;div&gt;&lt;h3&gt;Materials and Methods&lt;/h3&gt;&lt;div&gt;This study finally included 173 BC patients who underwent MRI between July 2023 and December 2024 and received pathological examination. All participants were evaluated with diffusion weighted imaging (DWI) and APTWI sequences. Group comparisons of clinical, MRI qualitative, and quantitative features, including apparent diffusion coefficient (ADC) and magnetization transfer ratio asymmetry (MTRasym) at 3.5 ppm, were conducted using one-way ANOVA, the Kruskal–Wallis test, or the χ² test as appropriate. Multivariate logistic regression identified independent predictors of HER2 status, with receiver operating characteristic analysis assessing models' performance for subclass discrimination.&lt;/div&gt;&lt;/div&gt;&lt;div&gt;This study finally included 173 BC patients who underwent MRI between July 2023 and December 2024 and received pathological examination. All participants were evaluated with diffusion weighted imaging (DWI) and APTWI sequences. Group comparisons of clinical, MRI qualitative, and quantitative features, including apparent diffusion coefficient (ADC) and magnetization transfer ratio asymmetry (MTRasym) at 3.5 ppm, were conducted using one-way ANOVA, the Kruskal–Wallis test, or the χ² test as appropriate. Multivariate logistic regression identified independent predictors of HER2 status, with receiver operating characteristic analysis assessing models' performance for subclass discrimination.&lt;/div&gt;&lt;div&gt;&lt;h3&gt;Results&lt;/h3&gt;&lt;div&gt;Clinically, estrogen receptor and progesterone receptor status, histological grade, Ki-67 index, and axillary lymph node differed significantly across HER2 subtypes (all P&lt;0.","PeriodicalId":50928,"journal":{"name":"Academic Radiology","volume":"33 10","pages":"Pages 4361-4373"},"PeriodicalIF":4.7,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148254036","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}