Machine Learning. Health最新文献

{"title":"Real-time quality feedback on Doppler data for community midwives using edge-AI.","authors":"Mohsen Motie-Shirazi, Sepideh Nikookar, Mohammad Ahmad, Alireza Rafiei, Reza Sameni, Peter Rohloff, Gari D Clifford, Nasim Katebi","doi":"10.1088/3049-477X/ae1bad","DOIUrl":"10.1088/3049-477X/ae1bad","url":null,"abstract":"<p><p>This study presents a technical framework for real-time fetal Doppler data quality assessment using deep learning and edge-AI, designed to improve data collection and support future clinical studies in low-resource settings. Integrated into a low-cost, edge-computing system co-designed with Indigenous midwives in rural Guatemala, our solution utilizes an Android phone for data acquisition and decision support. Retrospective analysis demonstrates the potential to detect fetal growth restriction, hypertension, and other pregnancy-related conditions using Doppler-based fetal cardiac signals. To ensure accurate assessments and provide immediate feedback, a real-time signal quality metric is essential. We analyzed two fetal Doppler datasets: 191 recordings, captured in rural Guatemala, for training and validation, and five captured in a German hospital (in Leipzig) for testing. The data were segmented into 3.75 s intervals, and categorized into five quality levels: good, poor, radiofrequency interference, talking, and silent. A deep neural network was trained on these segments, achieving a micro <math><mrow><mi>F</mi> <mn>1</mn></mrow> </math> score of 97.4% and a macro <math><mrow><mi>F</mi> <mn>1</mn></mrow> </math> score of 94.2%, with 99.2% accuracy for 'Good' quality in the Guatemala dataset, based on five-fold cross-validation. For the Leipzig dataset, the <math><mrow><mi>F</mi> <mn>1</mn></mrow> </math> score was 93.3% on 'Good' quality segments, demonstrating the model's ability to generalize across different datasets. By implementing the algorithm within an Android decision-support application in an mHealth framework, we have enabled real-time feedback during signal acquisition, improving data quality at the source. This scalable, edge mHealth solution offers significant potential to enhance maternal and fetal health monitoring in the Global South, contributing to global health efforts through the integration of mobile technology, AI, and healthcare.</p>","PeriodicalId":521035,"journal":{"name":"Machine Learning. Health","volume":"1 1","pages":"015014"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12625806/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145558869","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":"Incorporating physicians' contouring style into auto-segmentation of clinical target volume for post-operative prostate cancer radiotherapy using a language encoder.","authors":"Hengrui Zhao, Chien-Yi Liao, Daniel Yang, Steve Jiang, Dan Nguyen","doi":"10.1088/3049-477x/adf076","DOIUrl":"10.1088/3049-477x/adf076","url":null,"abstract":"<p><strong>Introduction: </strong>Post-operative radiotherapy for prostate cancer requires precise contouring of the clinical target volume (CTV) to account for microscopic disease that is invisible in the image. However, the absence of the prostate gland and unclear anatomical landmarks make this process challenging. In addition, the contouring style is greatly influenced by the physician's experience and preferences. This complicates automatic segmentation algorithms that rely on consistent data but also typically need large amounts data from many physicians to develop. This study presents a novel auto-segmentation method incorporating individual physicians' contouring styles to enhance accuracy and reduce manual workload.</p><p><strong>Methods: </strong>We develop Text-UNet, a deep learning-based model integrating a text encoder to encode physician-specific contouring styles into a latent vector. This information is combined with CT image features to guide segmentation. A dataset of 824 patients is divided into 699 training, 49 validation, and 76 testing data. The test set includes recent cases from four physicians, while the training set represents data from seven physicians.</p><p><strong>Results: </strong>Text-UNet achieves an average Dice score of 85.1%, outperforming the baseline UNet (82.0%) and state-of-the-art models. Incorporating physician-specific information improves segmentation consistency and reduces manual contouring effort, demonstrating its potential to enhance the auto-segmentation of post-operative radiotherapy for prostate cancer.</p><p><strong>Conclusion: </strong>By integrating physicians' contouring styles, Text-UNet addresses inter-physician variability, improving CTV auto-segmentation precision. This approach personalizes radiotherapy planning according to physicians and improves treatment efficiency.</p>","PeriodicalId":521035,"journal":{"name":"Machine Learning. Health","volume":"1 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12753031/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145879934","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":"Quantitative benchmarking of anomaly detection methods in digital pathology images.","authors":"Can Cui, Xindong Zheng, Ruining Deng, Quan Liu, Tianyuan Yao, Keith T Wilson, Lori A Coburn, Bennett A Landman, Haichun Yang, Yaohong Wang, Yuankai Huo","doi":"10.1088/3049-477X/ae0f9f","DOIUrl":"10.1088/3049-477X/ae0f9f","url":null,"abstract":"<p><p>Anomaly detection has been widely studied in the context of industrial defect inspection, with numerous methods developed to tackle a range of challenges. In digital pathology, anomaly detection holds significant potential for applications such as rare disease identification, artifact detection, and biomarker discovery. However, the unique characteristics of pathology images-such as large size, multi-scale structures, stain variability, and repetitive patterns-pose new challenges that current anomaly detection algorithms struggle to overcome. In this quantitative study, we benchmark 23 classical anomaly detection methods through extensive experiments. We systematically evaluate these approaches using five digital pathology datasets, including both real and synthetic cases. Our experiments investigate the influence of image scale, anomaly pattern types, and training epoch selection strategies on detection performance. The results provide a detailed comparison of each method's strengths and limitations, establishing a comprehensive benchmark to inform future research in anomaly detection for digital pathology. In addition, we review the current applications of anomaly detection algorithms in the field of pathology images. The code and simulation data will be publicly available at https://github.com/hrlblab/PathAnomalyDetect.</p>","PeriodicalId":521035,"journal":{"name":"Machine Learning. Health","volume":"1 1","pages":"010502"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12569974/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145411334","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":"Adaptive radiotherapy dose prediction on head and neck cancer patients with a 3D multi-headed U-Net deep learning architecture.","authors":"Hui-Ju Wang, Austen Maniscalco, David Sher, Mu-Han Lin, Steve Jiang, Dan Nguyen","doi":"10.1088/3049-477X/adfade","DOIUrl":"10.1088/3049-477X/adfade","url":null,"abstract":"<p><p>Online adaptive radiation therapy (ART) personalizes treatment plans by accounting for daily anatomical changes, requiring workflows distinct from conventional radiotherapy. Deep learning-based dose prediction models can enhance treatment planning efficiency by rapidly generating accuracy dose distributions, reducing manual trial-and-error and accelerating the overall workflow; however, most existing approaches overlook critical pre-treatment plan information-specifically, physician-defined clinical objectives tailored to individual patients. To address this limitation, we introduce the multi-headed U-Net (MHU-Net), a novel architecture that explicitly incorporates physician intent from pre-treatment plans to improve dose prediction accuracy in adaptive head and neck cancer treatments. Our dataset comprised 43 patients, each with pre-treatment plans, adaptive treatment plans, structure sets, and CT images. MHU-Net builds upon the widely adopted Stander U-Net architecture, extending it with a dual-head design: the primary head processes adaptive session contours and their corresponding signed distance maps, while the secondary head integrates pre-treatment contours, signed distance maps, and dose distributions. The features are merged within a primary U-Net framework to enhance dose prediction accuracy for adaptive treatment sessions. To evaluate the effectiveness of MHU-Net, we conducted a comparative analysis against U-Net. On average, MHU-Net reduced organ-at-risk dose prediction errors, achieving 1.78% lower maximum dose error and 1.22% lower mean dose error compared to U-Net. For the planning target volume, MHU-Net demonstrated significantly improved accuracy, with maximum and mean dose errors of 3.54  ±  2.75% and 1.07 ± 0.88%, respectively, outperforming U-Net's corresponding errors of 5.36 ± 4.19% and 2.76 ± 2.22% (<i>P</i> < 0.05). Taken together, these findings demonstrate that the proposed MHU-Met advances DL-based dose prediction for ART by effectively integrating both pre-treatment and adaptive session data. This approach facilitates the generation of dose distributions that more closely resemble the clinical ground truth, supporting personalization in ART planning and improving alignment with physician intent and treatment objectives.</p>","PeriodicalId":521035,"journal":{"name":"Machine Learning. Health","volume":"1 1","pages":"015008"},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12412777/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145014242","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":"Geometry-encoded deep learning (GeoDL) framework for real-time 3D dose verification for online adaptive radiotherapy.","authors":"Shunyu Yan, Austen Maniscalco, Biling Wang, Hengrui Zhao, Dan Nguyen, Steve Jiang, Chenyang Shen","doi":"10.1088/3049-477x/adcb4b","DOIUrl":"10.1088/3049-477x/adcb4b","url":null,"abstract":"<p><strong>Background: </strong>Accurate and efficient dose verification is vital for online adaptive radiotherapy (ART) while the patient is waiting on the treatment couch. Conventional methods are computationally intensive, limiting ART workflow efficiency and quality. A pilot study successfully illustrated the potential of geometry-encoded deep learning (GeoDL) algorithm for two-dimensional (2D) dose estimation on the central axial plane. This study investigates the feasibility of extending GeoDL for real-time dose verification for the entire three-dimensional (3D) volume of interest, fulfilling the unmet need in real clinical practice.</p><p><strong>Methods: </strong>The proposed GeoDL unifies the representation of fluence maps (FMs) and CT images by resolving the treatment delivery geometry of a linear accelerator (LINAC) to ease the burden of learning the complex domain transformation between FMs and CT image. A carefully designed 3D U-Net was trained on the integrated CT-FM volume to estimate the 3D dose distribution for the entire volume of interest, utilizing a combined loss function focusing on prediction errors in both high and low dose regions. The model was trained and validated on 281 prostate cancer cases, with 24 additional cases employed for independent testing.</p><p><strong>Results: </strong>The proposed GeoDL algorithm achieved highly accurate 3D dose estimation in almost real-time. The average γ passing rate ( <math><mn>3</mn> <mo>%</mo> <mo>/</mo> <mn>2</mn> <mi>m</mi> <mi>m</mi></math> , 10% threshold) for the estimated dose was 99.94% ± 0.15% on testing patients. The mean dose differences for the planning target volume (PTV) and organs at risk (OARs) were 0.93% ± 0.59% and 0.28% ± 0.32%, respectively. The average overall computational time for each testing case is ~35ms.</p><p><strong>Conclusion: </strong>This study demonstrates the feasibility of GeoDL in ultra-fast dose verification for ART, with strong performance in relevant dosimetric metrics across the entire volume of interest, positioning it as a valuable tool for real clinical use.</p>","PeriodicalId":521035,"journal":{"name":"Machine Learning. Health","volume":"1 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12753030/pdf/","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145879930","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}