BMC Medical Imaging最新文献

{"title":"Construction and validation of a unified 3D convolutional neural network framework for automated identification of cervical spinal cord injury based on CT imaging.","authors":"Deyan Ou, Yongxin Mo, Yu Huang, Yongqing Ye, Sen Lin, Linkun Pan, Rongdao Lin, Shengle Huang, Shixin Pan","doi":"10.1186/s12880-026-02626-3","DOIUrl":"https://doi.org/10.1186/s12880-026-02626-3","url":null,"abstract":"<p><strong>Objective: </strong>To construct and validate a standardized three-dimensional convolutional neural network (3D CNN) framework based on cervical spine CT imaging for automated identification of cervical spinal cord injury (CSCI), and to evaluate its stability across multicenter data and potential clinical applicability.</p><p><strong>Methods: </strong>Cervical spine CT images were retrospectively collected from two independent clinical centers, including cases with and without cervical spinal cord injury. All images were converted to NIfTI format through a unified preprocessing pipeline, and three-dimensional spinal cord regions of interest (ROIs) were delineated using ITK-SNAP to assess interobserver consistency. A standardized 3D CNN architecture was constructed, and the dataset was divided into training, internal validation, and external validation sets using a stratified random strategy. Model training was performed using the PyTorch framework. Classification performance across different datasets was systematically evaluated, and ROC curves, precision-recall (PR) curves, prediction probability distributions, and threshold sensitivity analyses were used to explore the model's clinical application characteristics.</p><p><strong>Results: </strong>The model demonstrated stable discriminative performance across the training, internal validation, and external validation sets, with AUC-ROC values of 0.943, 0.953, and 0.948, respectively, and AUC-PR values of 0.936, 0.939, and 0.943, respectively. Performance metrics showed minimal variation among datasets, with no evident overfitting or performance degradation. Prediction probability distributions indicated good separation between injured and non-injured samples. Threshold sensitivity analysis showed that the model could achieve performance trade-offs under different threshold settings, suggesting a degree of flexibility for clinical decision-making.</p><p><strong>Conclusion: </strong>Based on standardized data processing and network design, the 3D CNN effectively extracted latent spatial information associated with cervical spinal cord injury from CT images. The proposed framework demonstrated good stability and reproducibility across multicenter data and may serve as a preliminary research framework for CT-based CSCI risk evaluation, particularly in scenarios where MRI is unavailable or limited. Prospective validation is required before clinical deployment. It also provides a methodological reference for the standardized construction of deep learning models in medical imaging research.</p>","PeriodicalId":9020,"journal":{"name":"BMC Medical Imaging","volume":"26 1","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-10-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149285018","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Early regression of hepatic steatosis after bariatric surgery assessed by ultrasound including attenuation imaging.","authors":"Tatjana Hoffmann, Daniel Wulff, Sabrina Groß, Bariya Bajwa, Gunnar Blumenstock, André Mihaljevic, Nisar Malek, Lisa Minn","doi":"10.1186/s12880-026-02882-3","DOIUrl":"10.1186/s12880-026-02882-3","url":null,"abstract":"<p><strong>Background: </strong>Metabolic dysfunction-associated steatotic liver disease (MASLD) is associated with increased liver-related morbidity. Bariatric surgery provides substantial weight loss and metabolic improvement, yet data on ultrasound (US)-based monitoring remain limited. This study evaluated perioperative changes in hepatic steatosis using semiquantitative B-mode US and Attenuation Imaging (ATI), a quantitative US technique for fat quantification.</p><p><strong>Methods: </strong>In this prospective single-center study, patients undergoing bariatric surgery between February 2024 and July 2025 were included. B-mode US and ATI were performed preoperatively and at 3 and/or 6 months postoperatively. For overall analysis, the earliest available postoperative examination was defined as the early follow-up. Hepatic steatosis was assessed using B-mode US grading (S0-S3) and ATI grading (aS0-aS3). Body mass index (BMI), liver size, and attenuation coefficient (AC) values were recorded.</p><p><strong>Results: </strong>89 patients were included (mean age 42.8 years; 72% female). US examinations were performed at 3 months in 75.2% of patients and at 6 months in 51.7%. Of all patients, 26.9% attended both follow-up visits. At baseline, B-mode and ATI were strongly correlated (τ_b = 0.73, p < 0.001). At early follow-up, BMI decreased in all patients (p < 0.001). Grading of hepatic steatosis improved in 56.2% of patients by B-mode (p < 0.001) and in 68.5% by ATI (p < 0.001), while absolute ATI values decreased in 94.4%. Liver size decreased in 94.4%. In patients with both follow-up visits, largest reductions in ATI and liver size were observed at 3 months. Changes in B-mode and ATI grading were strongly correlated (τ_c = 0.65, p < 0.001).</p><p><strong>Conclusions: </strong>Regression of hepatic steatosis occurs early after bariatric surgery. ATI complements B-mode US by providing quantitative longitudinal assessment.</p>","PeriodicalId":9020,"journal":{"name":"BMC Medical Imaging","volume":"26 1","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-10-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13700481/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149262391","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Impact of reduced scan extent on tumor detection in ⁶⁸Ga-PSMA PET/CT for biochemical recurrence of prostate cancer.","authors":"Falk Gühne, Alexander Küppers, Christian Kühnel, Philipp Seifert, Martin Freesmeyer","doi":"10.1186/s12880-026-02940-w","DOIUrl":"10.1186/s12880-026-02940-w","url":null,"abstract":"<p><strong>Background: </strong>PSMA PET/CT plays an important role in biochemically recurrent prostate cancer (BCR-PCa). In general, PET/CT is performed as a limited or true whole-body examination. The majority of PET/CT scanners operate by measuring several bed positions or continuous motion with a limited axial field of view (FOV). Shortening the scan area thus results in a time saving, with unknown information loss regarding the detection of tumor manifestations.</p><p><strong>Methodology: </strong>This retrospective single-center study included 200 patients with BCR-PCa after initial curative therapy who underwent a ⁶⁸Ga-PSMA-PET/CT scan from vertex to toe (true whole-body). Using retrospective subdivision into hypothetical partial body measurement protocols, the undetected tumor manifestations were examined. Classic PET/CT protocols were used: vertex to mid-thigh, skull base to mid-thigh, as well as CT-equivalent examination areas: chest/abdomen/pelvis (CAP), abdomen/pelvis (AP) and pelvis.</p><p><strong>Results: </strong>Of 200 patients with a median PSA of 1.1 ng/mL, 128 (64%) showed at least one structural correlate of BCR-PCa in the true whole-body scan. The overall detection rate did not change for limited classic PET-protocols, or CAP measurements. CAP scans would not have detected all tumor manifestations in 16 patients (8%) and not all tumor-affected organ systems in six patients (3%), AP scans in 41 patients (21%) and 18 patients (9%), pelvic scans in 61 patients (31%) and 37 patients (19%), respectively. PSA may contribute to the prospective restriction of the scan area. Patients with PSA ≤ 0.75 ng/mL showed no diagnostic gain from whole-body imaging. This means that in 45% of patients in our cohort, a limitation to CAP scan would have been possible without losing information.</p><p><strong>Conclusions: </strong>For a resource-saving use of routine PSMA-PET/CT, a scan from the vertex to the mid-thigh seems sufficient for detecting relevant metastases. For patients with low PSA levels, a more focused protocol that includes only the CAP region can be considered.</p>","PeriodicalId":9020,"journal":{"name":"BMC Medical Imaging","volume":"26 1","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-10-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13685216/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149236875","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Unsupervised deep learning for inverse problems in computed tomography.","authors":"Laura Hellwege, Johann Christopher Engster, Moritz Schaar, Thorsten M Buzug, Maik Stille","doi":"10.1186/s12880-026-02794-2","DOIUrl":"https://doi.org/10.1186/s12880-026-02794-2","url":null,"abstract":"<p><p>Assume you encounter an inverse problem that shall be solved for a large number of data, but no ground-truth data is available. To emulate this, in this study we assume it is unknown how to solve the imaging problem of Computed Tomography. We introduce an unsupervised deep learning framework that leverages the inherent similarities between iterative reconstruction, Deep Image Prior (DIP), and unrolled optimization schemes. Our specific contribution is a training framework for amortized reconstruction: After training on a dataset without any image-domain ground truth, reconstruction of an unseen scan reduces to a single network forward pass. We demonstrate the feasibility of reconstructing images from measurement data by pure network inference, without additional gradient steps for unseen samples. Our method is evaluated on the two-dimensional 2DeteCT dataset. Within a controlled, geometry-matched benchmark, our reconstructions are competitive with, or better than, filtered back-projection, maximum-likelihood reconstruction, and a supervised network of identical architecture. Compared to a per-image DIP baseline, our method reaches similar quality while replacing the costly per-instance optimization with a single forward pass, yielding a speed-up of about four orders of magnitude. This makes it a promising candidate for time-critical imaging applications. Future work will address multi-dataset adaptability, counter-measures against over-smoothing, advanced uncertainty quantification, and further medical-imaging inverse problems.</p>","PeriodicalId":9020,"journal":{"name":"BMC Medical Imaging","volume":"26 1","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-09-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13628917/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149103873","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Comparative diagnostic performance of coronary CT angiography and invasive coronary angiography in patients with suspected coronary artery disease.","authors":"Arya Farzollahzadeh, Mohammad-Javad Shahsavari, Mehrnaz Mashoufi, Faraz Malek Bagali, Soheila Refahi","doi":"10.1186/s12880-026-02785-3","DOIUrl":"10.1186/s12880-026-02785-3","url":null,"abstract":"<p><strong>Background: </strong>Coronary artery disease (CAD) is a leading cause of morbidity and mortality worldwide. Although invasive coronary angiography (ICA) is the reference standard for diagnosing CAD, coronary computed tomography angiography (CCTA) has emerged as a valuable non-invasive alternative. This study compared the diagnostic performance of CCTA with ICA in patients with suspected CAD.</p><p><strong>Methods: </strong>In this descriptive cross-sectional study, 155 patients who underwent both CCTA and ICA were included. Coronary stenosis was classified as mild (1-49%), moderate (50-69%), or severe (≥ 70%). ICA served as the reference standard. Diagnostic performance was evaluated using sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy.</p><p><strong>Results: </strong>The mean age of the patients was 57.3 ± 9.15 years, and 67.7% were male. CCTA demonstrated a sensitivity of 96.0%, specificity of 33.3%, PPV of 85.7%, NPV of 66.7%, and an overall accuracy of 83.9%. The LAD showed the highest vessel-based sensitivity (87%), whereas the LCX had the lowest (60%). CCTA correctly identified all cases of moderate stenosis and detected 69% of severe stenoses.</p><p><strong>Conclusion: </strong>CCTA demonstrated high sensitivity for detecting CAD but relatively limited specificity in this high-risk population. These findings support its role as a useful non-invasive diagnostic tool, while ICA remains the reference standard for confirming coronary artery stenosis.</p>","PeriodicalId":9020,"journal":{"name":"BMC Medical Imaging","volume":"26 1","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-09-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13625413/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149079594","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"EEG functional connectivity measures for differentiating 6-month consciousness-related outcome after brain injury.","authors":"Sari Rahmawati Kusuma Dewi, Huan-Chieh Chen, Yu-Chuan Li, Hsuan-Chia Yang, Chih-Wei Huang, Lung Chan, Yong-Kwang Tu, Terry B J Kuo, Ming-Chin Lin","doi":"10.1186/s12880-026-02681-w","DOIUrl":"10.1186/s12880-026-02681-w","url":null,"abstract":"<p><strong>Background: </strong>Prognostication in patients with impaired consciousness after brain injury remains challenging because clinical assessments such as the Glasgow Coma Scale (GCS) may not fully capture underlying neurophysiological function. Electroencephalography (EEG) provides an accessible approach for assessing functional brain connectivity at the bedside. This study explored the association between EEG connectivity, clinical variables, and 6-month consciousness-related outcomes after brain injury.</p><p><strong>Methods: </strong>EEG was recorded from 111 hospitalized patients with impaired consciousness after brain injury. Baseline clinical status was assessed using the GCS. Patients were followed for 6 months and categorized according to command-following ability into good outcome and bad outcome groups. Functional connectivity was computed using weighted phase lag index (wPLI), debiased squared wPLI (wPLI²), and coherence. Clinical and EEG variables were evaluated using group comparisons, multivariable regression, receiver operating characteristic analysis, leave-one-out cross-validation, and sensitivity analyses.</p><p><strong>Results: </strong>Patients in the good outcome group were younger and had higher baseline GCS scores than those in the bad outcome group. Among EEG measures, alpha-band wPLI during the moving condition showed the strongest nominal group difference, with higher values in the good outcome group (nominal p = 0.006). This result did not survive false discovery rate correction across EEG connectivity variables (pFDR = 0.169). Alpha-band wPLI showed a nominal adjusted association with the 6-month consciousness-related outcome after adjustment for age and GCS, although the estimate was imprecise and attenuated in sensitivity analyses. In leave-one-out cross-validation, the clinical single-modality analysis using age and GCS achieved an AUC of 0.778, whereas the multimodality analysis combining age, GCS, and mean wPLI achieved an AUC of 0.733. The incremental improvement from adding EEG connectivity to clinical variables was not statistically significant.</p><p><strong>Conclusions: </strong>Alpha-band wPLI during task-based EEG showed an exploratory association with 6-month consciousness-related outcome after brain injury. However, this finding did not survive correction for multiple EEG comparisons and did not significantly improve discrimination beyond age and GCS in this cohort. Task-related EEG connectivity may serve as a candidate complementary marker, but larger multicenter studies with external validation are needed.</p>","PeriodicalId":9020,"journal":{"name":"BMC Medical Imaging","volume":"26 1","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-09-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13595776/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149015620","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"FDG-PET/CT-based treatment response predicts recurrence and overall survival in locally advanced cervical cancer.","authors":"Maria Markus, Hanna Sartor, Maria Bjurberg, David Minarik, Elin Trägårdh","doi":"10.1186/s12880-026-02772-8","DOIUrl":"10.1186/s12880-026-02772-8","url":null,"abstract":"<p><strong>Background: </strong>[18 F]-fluorodeoxyglucose (FDG) positron emission tomography combined with computed tomography (PET-CT) is used for staging and treatment response evaluation in patients with locally advanced cervical cancer. While the maximum standardised uptake value (SUVmax) is a commonly used PET parameter, volume-based parameters (metabolic tumour volume (MTV) and total lesion glycolysis (TLG)) may offer superior predictive value. This study aimed to investigate the predictive value of different treatment response outcomes assessed with PET-CT, using both standard and volume-based PET parameters.</p><p><strong>Methods: </strong>We retrospectively analyzed 133 patients with locally advanced cervical cancer treated with (chemo)radiotherapy who underwent PET-CT pre- and post-treatment. SUVmax, MTV and TLG were semi-automatically extracted from all scans, including tumour, lymph nodes and metastases. Treatment response was classified as complete metabolic response (CMR) (complete resolution of FDG uptake in all lesions), partial metabolic response (PMR) (> 25% reduction of SUVmax/MTV/TLG), progressive metabolic disease (PMD) (> 25% increase of SUVmax/MTV/TLG or appearance of new FDG-avid lesions) or stable disease (SD) (does not qualify for CMR, PMR or PMD). The three PET-parameters were analysed separately. PMR and SD were analysed as one group due to limited number of events. Associations between treatment response and overall survival (OS) as well as recurrence (binary variable) were analysed through Cox and logistic regression models, respectively.</p><p><strong>Results: </strong>For recurrence (binary variable), the odds ratio (OR) was 1.2 for patients with PMR/SD for SUVmax, MTV and TLG, while in the group of patients with PMD, it was 1.7 for SUVmax and 1.8 for MTV and TLG, using CMR as reference (i.e. for CMR, the OR is 1.0). The hazard ratio (HR) for OS was 4.0 for patients with PMR/SD and 16.0 for those with PMD, based on SUVmax; for MTV and TLG, HRs were 2.9 for patients with PMR/SD and 16.0 for those with PMD, using CMR as reference (HR 1.0). C-index from adjusted Cox models were comparable: 0.84 for SUVmax, and 0.86 for both MTV and TLG.</p><p><strong>Conclusion: </strong>FDG-PET/CT-based assessment of treatment response is a strong predictor of recurrence and overall survival in locally advanced cervical cancer. SUVmax, MTV, and TLG demonstrated similar predictive performance.</p>","PeriodicalId":9020,"journal":{"name":"BMC Medical Imaging","volume":"26 1","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13589335/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149001300","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"TransUNet-GradCAM: a hybrid transformer-U-Net with self-attention and explainable visualizations for foot ulcer segmentation.","authors":"Akwasi Asare, Mary Sagoe, Justice Williams Asare, Stephen E Moore","doi":"10.1186/s12880-026-02793-3","DOIUrl":"10.1186/s12880-026-02793-3","url":null,"abstract":"<p><p>Automated segmentation of diabetic foot ulcers (DFUs) supports clinical diagnosis, treatment planning, and longitudinal wound monitoring, but remains difficult owing to the heterogeneous appearance, irregular morphology, and cluttered backgrounds of ulcers in clinical photographs. Convolutional networks such as U-Net localise well but model long-range context poorly, whereas Vision Transformers capture global dependencies. We employ a hybrid ViT-bottleneck U-Net that combines a convolutional encoder-decoder with a Transformer bottleneck and attention-gated skip connections, and we emphasise that the contribution is a rigorously validated and explainable application rather than a new architecture. The model was trained on the public Foot Ulcer Segmentation Challenge (FUSeg) dataset with a hybrid Dice and cross-entropy loss, and all results are reported over five seeds as mean ± 95% confidence interval at a single fixed threshold. On the internal validation set it achieved a Dice of 0.8035 ± 0.0053 and an IoU of 0.7149 ± 0.0073 (HD95 = 19.74 px, ASSD = 6.12 px). A component-wise ablation showed that only the hybrid loss produced a statistically significant change in Dice (- 0.038, p < 0.001); the Transformer bottleneck, attention gates, and augmentation each had small, non-significant in-domain effects. External validation without retraining retained about 92% of internal Dice on the Advancing the Zenith of Healthcare (AZH) Wound Care Center cohort (Dice 0.7460, n = 278), while a small Medetec subset (n = 8) served only as a qualitative check, indicating partial rather than robust generalisation under domain shift. A quantitative explainability analysis found Grad-CAM more wound-localised (energy-in-mask 0.871 versus 0.102) but attention rollout significantly more faithful (p = 0.038, n = 200), showing the two are complementary. Predicted and expert wound areas agreed strongly (Pearson r = 0.944), and the model is lightweight (8.79 M parameters).</p>","PeriodicalId":9020,"journal":{"name":"BMC Medical Imaging","volume":"26 1","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13621505/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149073445","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Weakly supervised anatomical feature learning for cross-dataset ejection fraction estimation from echocardiography videos.","authors":"Adrian Krenzer, Viktoria Wieser, Tobias Friedetzki","doi":"10.1186/s12880-026-02781-7","DOIUrl":"10.1186/s12880-026-02781-7","url":null,"abstract":"<p><strong>Background: </strong>Ejection fraction (EF) is a central measure of cardiac function, but echocardiographic EF assessment remains reader-dependent and sensitive to acquisition quality. Deep learning can automate EF estimation, yet performance measured on a single development dataset may not transfer to data acquired under different imaging and annotation conventions. We investigated whether anatomically constrained weakly supervised learning improves cross-dataset EF estimation.</p><p><strong>Methods: </strong>We propose CAFEx, a contrastive-augmented feature extraction pipeline that combines left ventricular segmentation, echocardiography-specific augmentation, mask-derived anatomical features, and temporal EF regression. The model was trained on EchoNet-Dynamic using video-level EF labels with limited dense annotations for the anatomical component. External evaluation was performed on CAMUS after image harmonization and recalculation of an apical-four-chamber monoplane EF endpoint. Performance was compared with reproduced segmentation-based, direct video-regression, graph-based, transformer-based, and feature-extraction baselines using Dice score, mean absolute error, and coefficient of determination.</p><p><strong>Results: </strong>In the harmonized train-on-EchoNet/test-on-CAMUS benchmark, anatomically constrained models degraded less than direct video-regression baselines. CAFEx achieved 90.73% Dice on CAMUS and improved CAMUS EF prediction over the strongest reproduced feature-extraction baseline, increasing [Formula: see text] from 0.38 to 0.54 and reducing mean absolute EF error from 7.89 to 6.70 percentage points. Poor-quality videos and extreme EF values remained challenging.</p><p><strong>Conclusions: </strong>Weakly supervised EF estimation can benefit from an anatomical bottleneck learned with limited dense annotation. The findings support controlled external validation under explicit image and target harmonization, while further prospective validation, calibration, and quality-control mechanisms are needed before clinical deployment.</p>","PeriodicalId":9020,"journal":{"name":"BMC Medical Imaging","volume":"26 1","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-09-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13602555/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149036295","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Deep learning radiomics based on preoperative multiparametric MRI in predicting breast cancer recurrence risk.","authors":"Xinmiao Gong, Kepei Xu, Meiqi Hua, Yu Zhang, Hao Zheng, Sijia Fan, Chunjie Wang, Jidong Song, Changyu Zhou, Yangyang Bu, Kaiting Wang, Maosheng Xu, Ruixin Zhang","doi":"10.1186/s12880-026-02774-6","DOIUrl":"10.1186/s12880-026-02774-6","url":null,"abstract":"<p><strong>Background: </strong>Breast cancer treatments are often tailored to recurrence risk to improve outcomes, but reliable risk stratification methods are lacking.</p><p><strong>Purpose: </strong>To predict preoperative breast cancer recurrence risk through a tridimensional synergy model integrating breast MRI, radiomics, and deep learning.</p><p><strong>Study type: </strong>Retrospective.</p><p><strong>Population: </strong>428 female patients were randomly divided into training (n = 299, 52.5 ± 12.2 years) and internal validation (n = 129, 53.2 ± 11.5 years) sets, with an external validation set of 196 patients (51.3 ± 11.2 years).</p><p><strong>Field strength/sequence: </strong>Multiparametric 3T MRI included fat-suppressed T2-weighted (T2WI) spin-echo, axial diffusion-weighted imaging (DWI), and dynamic contrast-enhanced MRI (DCE-MRI) with one pre- and five post-contrast axial acquisitions.</p><p><strong>Assessment: </strong>We compared recurrence-free survival (RFS) prediction among DLR, DLC, and DLRC models, selected the optimal DLRC to stratify patients by risk, assessed RFS differences, and validated predictions, confirming effective risk stratification.</p><p><strong>Statistical tests: </strong>Continuous corrected chi-squared tests, one-way ANOVA, log-rank test. A two-tailed P< 0.05 was considered statistically significant.</p><p><strong>Results: </strong>The DL model showed predictive capability for 3-year RFS, with AUCs of 0.75 (95% CI, 0.65-0.84) in the training set, 0.74 (95% CI, 0.56-0.92) in the internal validation set, and 0.65 (95% CI, 0.53-0.78) in the external validation set. The DLR model achieved improved AUCs of 0.82 (95% CI: 0.74-0.90), 0.83 (95% CI: 0.73-0.93), and 0.67 (95% CI: 0.55-0.80) for predicting 3-year recurrence-free survival (RFS) in the training, internal validation, and external validation sets, respectively. The DLC model outperformed the DL model alone. The DLRC model achieved the best performance in the training and internal validation sets. Its predictive ability remained discernible but was attenuated in the external validation set, with AUCs of 0.95 (95% CI: 0.91-0.99), 0.89 (95% CI: 0.79-0.99), and 0.79 (95% CI: 0.67-0.91) across the respective datasets.</p><p><strong>Data conclusion: </strong>The DLRC model effectively predicted and stratified breast cancer recurrence risk by integrating deep learning, radiomics, and clinicopathological features.</p><p><strong>Evidence level: </strong>3.</p><p><strong>Technical efficacy: </strong>Stage 5.</p>","PeriodicalId":9020,"journal":{"name":"BMC Medical Imaging","volume":"26 1","pages":""},"PeriodicalIF":3.9,"publicationDate":"2026-09-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13599116/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149029722","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"医学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}