Journal of Medical Imaging最新文献

{"title":"Unsupervised anomaly detection with deep generative models: an experimental analysis of model variability and mitigation strategies.","authors":"Maëlys Solal, Pascaline André, Ninon Burgos, For The Alzheimer's Disease Neuroimaging Initiative Group","doi":"10.1117/1.JMI.13.4.044003","DOIUrl":"10.1117/1.JMI.13.4.044003","url":null,"abstract":"<p><strong>Purpose: </strong>Unsupervised anomaly detection allows identifying anomalies from unlabeled data, making it useful for neuroimaging analysis and computer-aided diagnosis. Given an individual's scan, we use a generative model to construct a subject-specific image of healthy appearance and compare both images with identify anomalies. Such approach has drawbacks as the reconstructions are imperfect, and model variability is not taken into account.</p><p><strong>Approach: </strong>We study model variability arising from using different random seeds during training and explore strategies to mitigate the effect of unwanted reconstruction errors and variability. The strategies include model ensembling, anomaly map normalization, and anomaly map designs based on single or multiple pseudo-healthy reconstructions. We compare these approaches in the context of dementia-related anomalies on 3D FDG PET from ADNI using variational autoencoder models.</p><p><strong>Results: </strong>Our experiments suggest that variance between models can be reduced by aggregating their reconstructions in a <math><mrow><mi>Z</mi></mrow> </math> -score based anomaly map. This strategy is highly effective, but computationally expensive, as it requires training several instances of the model. An alternative strategy is to normalize the anomaly map using statistics computed from a healthy validation set. We show that this normalization strategy substantially reduces performance variability across models and even increases anomaly detection performance in certain cases.</p><p><strong>Conclusion: </strong>We highlight a largely overlooked source of variability in deep generative models, their random seed, which can lead to substantially different anomaly detection performance and biased evaluation. Our experimental analysis emphasizes that it is crucial to design robust anomaly maps and to mitigate the impact of randomness-induced variability, by accounting for model reconstruction errors for instance using <math><mrow><mi>Z</mi></mrow> </math> -score ensembling, or healthy-control normalization, to support more stable and trustworthy deployment in real-world clinical practice.</p>","PeriodicalId":47707,"journal":{"name":"Journal of Medical Imaging","volume":"13 4","pages":"044003"},"PeriodicalIF":2.3,"publicationDate":"2026-07-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13537101/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148882143","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":"Methodological considerations for evaluating deep learning segmentation models in digital pathology whole-slide images.","authors":"Arian Arab, Victor Garcia, Seyed Kahaki, Mart van Rijthoven, Roberto Salgado, Brandon D Gallas, Nicholas Petrick, Francesco Ciompi, Weijie Chen","doi":"10.1117/1.JMI.13.3.037501","DOIUrl":"https://doi.org/10.1117/1.JMI.13.3.037501","url":null,"abstract":"<p><strong>Purpose: </strong>Automated whole-slide image (WSI) analysis, specifically applications of deep learning (DL)-based algorithms, has been enabling automated detection, classification, segmentation, and prognosis for various diseases. Performance evaluation plays an important role in the success of these complex big-data-based technologies. Our purpose is to conduct a performance evaluation of DL segmentation models applied to a breast cancer WSI dataset provided by the Tumor InfiltratinG lymphocytes in breast cancER challenge and investigate methodological issues in the assessment of WSI segmentation models.</p><p><strong>Approach: </strong>We evaluated the performance of DL models in the segmentation of tumoral and stromal regions and the effect of color normalization on improving the performance of these models when the training and testing data are from different sources. One important issue is the aggregation of image segmentation performance when the reference standard includes annotations only from selected regions of interest (ROIs). We introduced three different methods for aggregating performance based on different units of analysis (pixels, ROIs, and slides) and a bootstrap method to estimate the variance of the performance results at the slide level.</p><p><strong>Results: </strong>We found that using different units of analysis can produce not just different mean performance estimates but also different levels of uncertainty. Our results also showed that color normalization significantly improved DL model performance when the training and testing data are from different sources.</p><p><strong>Conclusions: </strong>Our study demonstrates the importance of image acquisition, study design, and statistical analysis methods used in the performance evaluation of computational pathology applications.</p>","PeriodicalId":47707,"journal":{"name":"Journal of Medical Imaging","volume":"13 3","pages":"037501"},"PeriodicalIF":2.3,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13267111/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148266613","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":"Literature Reviews After AI.","authors":"Sam Bogdanov, Andre Teixeira da Silva Hucke, Bennett A Landman","doi":"10.1117/1.JMI.13.3.030101","DOIUrl":"https://doi.org/10.1117/1.JMI.13.3.030101","url":null,"abstract":"<p><p>This editorial revisits the enduring role of literature reviews in scientific training and communication, emphasizing that their value lies in expert synthesis rather than simple aggregation of sources. In an era of AI-assisted discovery and summarization, the authors argue that peer-reviewed reviews must provide deeper perspective on how fields evolve and how evidence interconnects. They outline updated expectations for review and perspective articles, stressing that human judgment remains essential for interpreting and validating the scientific record.</p>","PeriodicalId":47707,"journal":{"name":"Journal of Medical Imaging","volume":"13 3","pages":"030101"},"PeriodicalIF":2.3,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13308937/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148346900","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":"DDARes-U²Net: a dual-decoder adversarial residual U²Net algorithm for segmentation of COVID-19 pneumonia lesions.","authors":"Xiao Li, Fujiao Ju, Yifei Xu, Haoran Chen","doi":"10.1117/1.JMI.13.3.034002","DOIUrl":"https://doi.org/10.1117/1.JMI.13.3.034002","url":null,"abstract":"<p><strong>Purpose: </strong>We aim to overcome the remaining bottlenecks in COVID-19 lesion segmentation from chest CT-namely, blurred lesion boundaries, false-positive responses from vessels or trachea, and the extreme variability of lesion shape, size, and location-by developing a dual-decoder adversarial residual <math> <mrow> <msup><mrow><mi>U</mi></mrow> <mrow><mn>2</mn></mrow> </msup> <mtext>Net</mtext></mrow> </math> algorithm.</p><p><strong>Approach: </strong>DDARes- <math> <mrow> <msup><mrow><mi>U</mi></mrow> <mrow><mn>2</mn></mrow> </msup> <mtext>Net</mtext></mrow> </math> , an encoder-dual-decoder architecture, simultaneously performs both lung parenchyma segmentation and lesion segmentation within a unified network. The dual-decoder design incorporates a lung-parenchyma decoder that guides the model to focus specifically on the lung region, eliminating the information loss caused by traditional presegmentation approaches and markedly reducing computational overhead. Furthermore, an edge-loss function is introduced during training to enhance the network's ability to capture fine-grained lesion boundaries. Comparative experiments are conducted under two settings: within-dataset evaluation and cross-dataset evaluation. Ablation studies are performed to validate the contribution of each architectural component.</p><p><strong>Results: </strong>The proposed DDARes- <math> <mrow> <msup><mrow><mi>U</mi></mrow> <mrow><mn>2</mn></mrow> </msup> <mtext>Net</mtext></mrow> </math> model, with a relatively modest parameter count of 30.74 M, achieves superior generalization performance and segmentation accuracy compared with state-of-the-art methods. Across both within-center and cross-center experiments on five CT datasets (I to V), DDARes- <math> <mrow> <msup><mrow><mi>U</mi></mrow> <mrow><mn>2</mn></mrow> </msup> <mtext>Net</mtext></mrow> </math> consistently ranks first for Dice and IoU. The largest gain is observed when the model trained on Dataset IV is tested on Dataset II, delivering <math><mrow><mo>+</mo> <mn>3.41</mn></mrow> </math> pp Dice versus ARes- <math> <mrow><msup><mi>U</mi> <mn>2</mn></msup> <mi>Net</mi> <mo>+</mo> <mo>+</mo></mrow> </math> and <math><mrow><mo>+</mo> <mn>5.51</mn></mrow> </math> pp versus ARes- <math> <mrow> <msup><mrow><mi>U</mi></mrow> <mrow><mn>2</mn></mrow> </msup> <mtext>Net</mtext></mrow> </math> , confirming its superior generalizability and clinical applicability.</p><p><strong>Conclusions: </strong>The unified dual-decoder framework with explicit edge-aware supervision provides accurate, robust, and generalizable COVID-19 lesion segmentation, achieving the highest Dice and IoU in both intradataset and cross-center tests on five independent CT cohorts.</p>","PeriodicalId":47707,"journal":{"name":"Journal of Medical Imaging","volume":"13 3","pages":"034002"},"PeriodicalIF":2.3,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13278900/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148297035","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":"Illustration of transfer learning from breast cancer detection to risk prediction: adaptation to local data and local objectives.","authors":"Tobias Wagner, Zan Klanecek, Yao-Kuan Wang, Lesley Cockmartin, Nicholas W Marshall, Andrej Studen, Robert Jeraj, Hilde Bosmans","doi":"10.1117/1.JMI.13.3.034504","DOIUrl":"https://doi.org/10.1117/1.JMI.13.3.034504","url":null,"abstract":"<p><strong>Purpose: </strong>We aim to investigate whether a breast cancer risk model can be trained with transfer learning from a breast cancer detection model.</p><p><strong>Approach: </strong>An existing open-source breast cancer detection model was used to extract the latent space representation of a local dataset composed of images from 17,878 breast cancer screening participants. An autoencoder network was used for dimensionality reduction. A final model training step mapped the reduced representation to 4-year breast cancer risk. Two experimental studies were performed: first, a local dataset was used to investigate the impact of transfer learning to local data. Second, the training was executed for breast cancer risk by excluding cases with obvious signs of cancer in the model training process, to avoid the influence of cancer signs. In addition, the Mirai model was evaluated.</p><p><strong>Results: </strong>The open-source model achieved a baseline performance of 0.72 with 95%CI [0.66, 0.78] for 4-year risk prediction when applied to local test cases of breast cancer risk data. This performance was improved after training on local breast cancer detection data to 0.75 with 95%CI [0.69, 0.81], although not statistically significant ( <math><mrow><mi>p</mi> <mo>=</mo> <mn>0.09</mn></mrow> </math> ). Further training with local breast cancer risk data achieved an area under the curve (AUC) of 0.77 with 95%CI [0.71, 0.82], which is higher than the original model ( <math><mrow><mi>p</mi> <mo><</mo> <mn>0.01</mn></mrow> </math> ). The Mirai model demonstrated a lower performance after fine-tuning ( <math><mrow><mi>p</mi> <mo><</mo> <mn>0.01</mn></mrow> </math> ).</p><p><strong>Conclusions: </strong>Transfer learning enabled the creation of a breast cancer risk prediction model from a previously developed breast cancer detection model, demonstrating that a publicly available model can be adapted to local clinical needs.</p>","PeriodicalId":47707,"journal":{"name":"Journal of Medical Imaging","volume":"13 3","pages":"034504"},"PeriodicalIF":2.3,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13309842/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148346842","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":"CLEAR-AI: confounder-aware learning for equitable and accurate reasoning in AI for diagnosis.","authors":"Vedant Joshi, Ramon Correa, Avisha Das, Judy Gichoya, Imon Banerjee","doi":"10.1117/1.JMI.13.3.034503","DOIUrl":"https://doi.org/10.1117/1.JMI.13.3.034503","url":null,"abstract":"<p><strong>Purpose: </strong>A critical challenge impeding the deployment of artificial intelligence (AI) models in healthcare lies in implicit bias against multiple correlated sensitive attributes.</p><p><strong>Approach: </strong>We developed a multibranch adversarial debiasing approach that can debias a multilabel diagnostic model simultaneously for correlating confounding factors using a dynamic weighted gradient reversal technique to reduce disparities. The proposed methodology was evaluated on chest X-ray imaging data-trained on CheXpert and evaluated on two external datasets [MIMIC-CXR (Beth Israel, Massachusetts, United States) and Emory Healthcare (Atlanta, Georgia, United States)].</p><p><strong>Results: </strong>The disease classification performance of the proposed debiased method significantly overlaps with the baselines, indicating no drop in the task performance. We found the debiased model to reduce TPR and FPR disparities across multifactor subgroups [age, race, and support device(s)] while maintaining overall task performance.</p><p><strong>Conclusions: </strong>Although we focused on three specific confounders, the proposed adversarial debiasing framework readily extends to account for an arbitrary number of sensitive variables. The findings highlight the promising potential of adversarial training techniques to enhance fairness and trustworthiness in the deployment of AI models in diverse healthcare settings.</p>","PeriodicalId":47707,"journal":{"name":"Journal of Medical Imaging","volume":"13 3","pages":"034503"},"PeriodicalIF":1.7,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13198178/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148018206","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":"Pediatric-specific computer-aided detection of lung nodules in computed tomography scans.","authors":"Russell C Hardie, Dylan Flaute, Barath N Narayanan, Andrew T Trout, Jonathan R Dillman, Pradipta Debnath, Alexandra Glenn, Aki A Tanimoto","doi":"10.1117/1.JMI.13.3.034502","DOIUrl":"10.1117/1.JMI.13.3.034502","url":null,"abstract":"<p><strong>Purpose: </strong>In pediatric patients with cancer, the presence of small lung nodules and micronodules on computed tomography (CT) scans can be clinically significant and indicative of metastatic cancer. Lung nodule computer-aided detection (CAD) systems trained with larger adult nodules have been shown to be less sensitive to smaller nodules in pediatric patients. To our knowledge, no pediatric-trained lung nodule CAD systems have been evaluated in the literature. To help address this gap, we compare the performance of two CAD systems, both trained on adult data and separately with pediatric data.</p><p><strong>Approach: </strong>We developed a lung nodule CAD system, Deep FlyerScan, and compared it with the Medical Open Network for Artificial Intelligence (MONAI) lung nodule detection system. Both systems are initially trained using adult data and then each is fine-tuned with pediatric data. Performance is evaluated using a retrospective pediatric dataset comprising 75 CT scans. Reference standard nodules have been provided by multiradiologist review from an earlier study.</p><p><strong>Results: </strong>The pediatric fine-tuned Deep FlyerScan and MONAI systems detected 74.8% (404/540) and 64.6% (349/540) of pediatric nodules at 5 false positives per scan, respectively. This compares with 62.0% (335/540) and 55.0% (297/540), respectively, for their counterparts trained using only adult data.</p><p><strong>Conclusions: </strong>The two CAD systems studied showed increased sensitivity in detecting lung nodules in pediatric CT scans when fine-tuned using pediatric data compared with training using only adult data. The improvement may be related to reference standard nodule size differences between the adult and pediatric datasets.</p>","PeriodicalId":47707,"journal":{"name":"Journal of Medical Imaging","volume":"13 3","pages":"034502"},"PeriodicalIF":2.3,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13167107/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147934562","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":"Deep-learning-based multiple organ segmentation for CT scout images: applications to automatic CT planning.","authors":"Kaylee W Fang, Sen Wang, Maria Jose Medrano, Liyan Sun, Grant M Stevens, Justin R Tse, Adam S Wang","doi":"10.1117/1.JMI.13.3.033501","DOIUrl":"https://doi.org/10.1117/1.JMI.13.3.033501","url":null,"abstract":"<p><strong>Purpose: </strong>Computed tomography (CT) scout images are used in CT planning to set the anatomic scan range and optimize the radiation dose. Manual exam planning is highly variable and contributes to excess radiation dose delivered to patients. We present a multi-organ segmentation system for CT scout images with a direct application for automatic scan range prediction, an important component of CT planning.</p><p><strong>Approach: </strong>Variants of a U-Net model with a ResNet-50 encoder were trained to segment 35 distinct anatomic structures from frontal and lateral scout images and evaluated on their performance. Organ boundaries generated by the trained models were used to automatically set the scan range for chest-abdomen-pelvis and abdomen-pelvis CT protocols from frontal and lateral scout images.</p><p><strong>Results: </strong>The best-performing models produced organ segmentations with mean Dice scores of <math><mrow><mn>0.823</mn> <mo>±</mo> <mn>0.014</mn></mrow> </math> for the frontal scout and <math><mrow><mn>0.784</mn> <mo>±</mo> <mn>0.013</mn></mrow> </math> for the lateral scout. Predicted scan ranges covered ground-truth anatomy for 87% to 100% of test cases, reduced variability in six out of eight tested scan limits, and produced shorter median scan lengths for three out of four scout-to-protocol prediction tasks.</p><p><strong>Conclusions: </strong>The presented models effectively segment a wide variety of organs from frontal and lateral scout images and demonstrate promising clinical applications to improved scan range prediction and automatic CT planning.</p>","PeriodicalId":47707,"journal":{"name":"Journal of Medical Imaging","volume":"13 3","pages":"033501"},"PeriodicalIF":1.7,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13210884/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148043873","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":"High-speed optical tracking and augmented reality platform for image-guided interventions.","authors":"Nati Nawawithan, James Yu, Kelden Pruitt, Baowei Fei","doi":"10.1117/1.JMI.13.3.035001","DOIUrl":"10.1117/1.JMI.13.3.035001","url":null,"abstract":"<p><strong>Purpose: </strong>During interventional procedures, clinicians need to mentally register anatomical information from preoperative cross-sectional images onto the patient's body to envision the location of subsurface targets and critical structures. This study aims to address this challenge by developing an augmented reality (AR)-based tracking system that can leverage cross-sectional images to provide real-time three-dimensional (3D) visualization of lesion targets and enable precise surgical procedures.</p><p><strong>Approach: </strong>Our AR platform combines a customized high-speed, real-time, optical tracking system, a holographic display device, a computer workstation, and a graphical user interface. The system displays holograms of virtual models of the target organ and surgical tools, as well as the navigation path. To validate our AR platform, we measured target registration errors (TRE) across two different types of interventional procedures using our high-speed optical tracking system. To conduct these experiments, we performed laparoscopic procedures and prostate biopsies on customized phantoms.</p><p><strong>Results: </strong>The integrated high-speed tracking and AR platform was applied to laparoscopic and biopsy procedures. The average TRE was <math><mrow><mn>4.17</mn> <mo>±</mo> <mn>1.63</mn> <mtext>  </mtext> <mi>mm</mi></mrow> </math> for laparoscopic procedures and <math><mrow><mn>2.89</mn> <mo>±</mo> <mn>0.84</mn> <mtext>  </mtext> <mi>mm</mi></mrow> </math> for prostate biopsy.</p><p><strong>Conclusions: </strong>An augmented reality platform with a high-speed precision optical tracking system was developed for interventional procedures. The AR platform has been demonstrated for potential applications in prostate laparoscopic and biopsy procedures and can be expanded to other interventional procedures.</p>","PeriodicalId":47707,"journal":{"name":"Journal of Medical Imaging","volume":"13 3","pages":"035001"},"PeriodicalIF":2.3,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13273055/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148279255","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":"Multi-institutional classification of fibroglandular tissue and background parenchymal enhancement in breast MRI using deep learning.","authors":"Kai Geissler, Hans Meine, Hendrik Laue, Robert Grimm, Heinrich von Busch, Susanne Diekmann, Dieter Szolar, Sabine Ohlmeyer, Edyta Szurowska, Uwe Fischer, Nachiko Uchiyama, Kazuki Oyama, Noemi Schmidt, Markus Wenzel","doi":"10.1117/1.JMI.13.3.034501","DOIUrl":"https://doi.org/10.1117/1.JMI.13.3.034501","url":null,"abstract":"<p><strong>Purpose: </strong>The amount of fibroglandular tissue (FGT) and background parenchymal enhancement (BPE) are evaluated as part of the BI-RADS reporting standard for the diagnosis of breast cancer in magnetic resonance imaging (MRI). BPE is associated with breast cancer risk and affects the diagnostic accuracy of breast MRI readings but is subject to very high inter-reader variability.</p><p><strong>Approach: </strong>We developed a fully automatic observer-independent method to measure the amount of FGT and BPE based on breast, FGT, and lesion masks, which classifies each into four classes based on thresholds applied to those scores. This retrospective study involved 2840 patients from seven institutions for whom T1-weighted dynamic contrast-enhanced breast MRI sequences with and without fat suppression were acquired using 1.5 or 3T scanners. FGT and BPE assessments for model development and evaluation were performed according to the BI-RADS guidelines by radiologists from the respective institutions and by a radiographer if the assessments were missing.</p><p><strong>Results: </strong>The performance of the models was compared using four-class accuracy and a paired <i>t</i>-test ( <math><mrow><mi>α</mi> <mo>=</mo> <mn>0.05</mn></mrow> </math> ). Our automatic classification of FGT and BPE in a multi-institutional setting achieved a four-class accuracy of 0.535 for FGT and 0.504 for BPE ( <math><mrow><mi>random</mi> <mo>=</mo> <mn>0.25</mn></mrow> </math> ). The calibration of the classification thresholds for each site individually resulted in a significant improvement, reaching an accuracy of 0.669 ( <math><mrow><mi>p</mi> <mo><</mo> <mn>0.05</mn></mrow> </math> ) for FGT and 0.565 ( <math><mrow><mi>p</mi> <mo><</mo> <mn>0.05</mn></mrow> </math> ) for BPE. The area under curve of receiver operating characterstic (ROC-AUC) to classify FGT and BPE into low versus high is 0.883 for FGT and 0.757 for BPE when computed across all sites. The averaged per-site AUCs are 0.926 for FGT and 0.810 for BPE.</p><p><strong>Conclusion: </strong>FGT and BPE classification perform significantly better when classification thresholds are selected per clinical site.</p>","PeriodicalId":47707,"journal":{"name":"Journal of Medical Imaging","volume":"13 3","pages":"034501"},"PeriodicalIF":1.7,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC13168875/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147943391","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}