Robbe D’hondt , Klest Dedja , Sofie Aerts , Bart Van Wijmeersch , Tomas Kalincik , Stephen Reddel , Eva Kubala Havrdova , Alessandra Lugaresi , Bianca Weinstock-Guttman , Saloua Mrabet , Patrice Lalive , Allan G. Kermode , Serkan Ozakbas , Francesco Patti , Alexandre Prat , Valentina Tomassini , Izanne Roos , Raed Alroughani , Oliver Gerlach , Samia J. Khoury , Celine Vens
{"title":"Explainable time-to-progression predictions in multiple sclerosis","authors":"Robbe D’hondt , Klest Dedja , Sofie Aerts , Bart Van Wijmeersch , Tomas Kalincik , Stephen Reddel , Eva Kubala Havrdova , Alessandra Lugaresi , Bianca Weinstock-Guttman , Saloua Mrabet , Patrice Lalive , Allan G. Kermode , Serkan Ozakbas , Francesco Patti , Alexandre Prat , Valentina Tomassini , Izanne Roos , Raed Alroughani , Oliver Gerlach , Samia J. Khoury , Celine Vens","doi":"10.1016/j.cmpb.2025.108624","DOIUrl":"10.1016/j.cmpb.2025.108624","url":null,"abstract":"<div><h3>Background:</h3><div>Prognostic machine learning research in multiple sclerosis has been mainly focusing on black-box models predicting whether a patients’ disability will progress in a fixed number of years. However, as this is a binary yes/no question, it cannot take individual disease severity into account. Therefore, in this work we propose to model the time to disease progression instead. Additionally, we use explainable machine learning techniques to make the model outputs more interpretable.</div></div><div><h3>Methods:</h3><div>A preprocessed subset of 29,201 patients of the international data registry MSBase was used. Disability was assessed in terms of the Expanded Disability Status Scale (EDSS). We predict the time to significant and confirmed disability progression using random survival forests, a machine learning model for survival analysis. Performance is evaluated on a time-dependent area under the receiver operating characteristic and the precision–recall curves. Importantly, predictions are then explained using SHAP and Bellatrex, two explainability toolboxes, and lead to both global (population-wide) as well as local (patient visit-specific) insights.</div></div><div><h3>Results:</h3><div>On the task of predicting progression in 2 years, the random survival forest achieves state-of-the-art performance, comparable to previous work employing a random forest. However, here the random survival forest has the added advantage of being able to predict progression over a longer time horizon, with AUROC <span><math><mrow><mo>></mo><mn>60</mn><mtext>%</mtext></mrow></math></span> for the first 10 years after baseline. Explainability techniques further validated the model by extracting clinically valid insights from the predictions made by the model. For example, a clear decline in the per-visit probability of progression is observed in more recent years since 2012, likely reflecting globally increasing use of more effective MS therapies.</div></div><div><h3>Conclusion:</h3><div>The binary classification models found in the literature can be extended to a time-to-event setting without loss of performance, thus allowing a more comprehensive prediction of patient prognosis. Furthermore, explainability techniques proved to be key to reach a better understanding of the model and increase validation of its behaviour.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"263 ","pages":"Article 108624"},"PeriodicalIF":4.9,"publicationDate":"2025-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143418849","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}
Nannan Cao , Qilin Li , Kangkang Sun , Heng Zhang , Jiangyi Ding , Ziyi Wang , Wei Chen , Liugang Gao , Jiawei Sun , Kai Xie , Xinye Ni
{"title":"MBST-Driven 4D-CBCT reconstruction: Leveraging swin transformer and masking for robust performance","authors":"Nannan Cao , Qilin Li , Kangkang Sun , Heng Zhang , Jiangyi Ding , Ziyi Wang , Wei Chen , Liugang Gao , Jiawei Sun , Kai Xie , Xinye Ni","doi":"10.1016/j.cmpb.2025.108637","DOIUrl":"10.1016/j.cmpb.2025.108637","url":null,"abstract":"<div><h3>Objective</h3><div>This research developed an innovative Mask-based Swin Transformer network (MBST) to enhance the quality of 4D cone-beam computed tomography (4D-CBCT) reconstruction. The network is trained on 4D-CBCT reconstructed under limited scanning conditions, enabling its application to a broad range of 4D-CBCT reconstruction scenarios, including those with high scanning speeds.</div></div><div><h3>Methods</h3><div>4D imaging data from 20 patients with thoracic tumors were used to train and evaluate the deep learning model. 15 cases were used for training, and 5 cases were employed for simulation testing. The Feldkamp–Davis–Kress algorithm was employed to simulate 4D-CBCT from downsampled 4D-CT data to mitigate the uncertainties associated with respiratory motion between treatment fractions, and the 4D-CT data served as the ground truth for training. The study reconstructed 4D-CBCT images under 11 different scanning intervals including full angle acquisition at 1°, 2°, 3°, 4°, 5°, 6°, 12°, 18°, 24° intervals, and 1/3 full angles acquisition at 5°, 10° inrevals respectively for capturing 4D-CBCT projections. The test results were quantitatively evaluated using the structural similarity index measure (SSIM), peak signal-to-noise ratio (PSNR), mean error (ME), and mean absolute error (MAE), and image quality was qualitatively assessed. Real clinical patients who were not included in the training were tested to evaluate the network's ability to generalize. Moreover, the proposed method was compared with other deep learning approaches, and statistical analyses were performed.</div></div><div><h3>Results</h3><div>Simulation data assessment revealed that with small projection acquisition interval, such as the 4°interval, the 4D-CBCT images optimized by MBST showed a considerable improvement over the original 4D-CBCT images in terms of SSIM (42.3% increase) and PSNR (10.8 dB increase), and the ME and MAE values approached 0. The improvements were statistically significant (<em>P</em> < 0.001). Compared with other deep learning methods, MBST demonstrated superior performance with improvements of 1.4% in SSIM and 1.21 dB in PSNR and a reduction of 0.94 in MAE. With large projection intervals, such as the 24°interval, MBST outperformed other deep learning methods. Specifically, its SSIM, PSNR, and MAE increased by 3.8%, 0.81 dB, and 10.34, respectively, compared with those of other deep learning methods, and the improvements were statistically significant (P < 0.01). In addition, MBST could reconstruct bone tissue and optimize the quality of 4D-CBCT images even when the number of projections was small (12°, 18°, 24°intervals). Clinical data evaluation revealed that after optimization by MBST, the SSIM, PSNR, ME, and MAE of 4D-CBCT compared with those of 4D-CT registration improved from the original 22.8%, 15.49 dB, −345.5, and 432.2 to 81.5%, 27.93 dB, −53.79, and 73.77, respectively. Moreover, MBST exhibited the most pronounced impro","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"262 ","pages":"Article 108637"},"PeriodicalIF":4.9,"publicationDate":"2025-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143379300","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}
Ming-Feng Tsai , Yu-Chang Chu , Wen-Teng Yao , Chia-Meng Yu , Yu-Fan Chen , Shu-Tien Huang , Liong-Rung Liu , Lang-Hua Chiu , Yueh-Hung Lin , Chin-Yi Yang , Kung-Chen Ho , Chieh-Ming Yu , Wen-Chen Huang , Sheng-Yun Ou , Kwang-Yi Tung , Fei-Hung Hung , Hung-Wen Chiu
{"title":"Deep-learning-based diagnosis framework for ankle-brachial index defined peripheral arterial disease of lower extremity wound: Comparison with physicians","authors":"Ming-Feng Tsai , Yu-Chang Chu , Wen-Teng Yao , Chia-Meng Yu , Yu-Fan Chen , Shu-Tien Huang , Liong-Rung Liu , Lang-Hua Chiu , Yueh-Hung Lin , Chin-Yi Yang , Kung-Chen Ho , Chieh-Ming Yu , Wen-Chen Huang , Sheng-Yun Ou , Kwang-Yi Tung , Fei-Hung Hung , Hung-Wen Chiu","doi":"10.1016/j.cmpb.2025.108654","DOIUrl":"10.1016/j.cmpb.2025.108654","url":null,"abstract":"<div><h3>Background and Objective</h3><div>Few studies have evaluated peripheral artery disease (PAD) in patients with lower extremity wounds by a convolutional neural network (CNN)-based deep learning algorithm. We aimed to establish a framework for PAD detection, peripheral arterial occlusive disease (PAOD) detection, and PAD classification in patients with lower extremity wounds by the AlexNet, GoogleNet, and ResNet101V2 algorithms.</div></div><div><h3>Methods</h3><div>Our proposed framework was based on a CNN-based AlexNet, GoogleNet, or ResNet 101V2 model devoted to performing optimized detection and classification of PAD in patients with lower extremity wounds. We also evaluated the performance of the plastic and reconstructive surgeons (PRS) and general practitioner (GP).</div></div><div><h3>Results</h3><div>Compared to the performance of AlexNet or GoogleNet, a slight increase in ResNet101V2-based performance of PAD detection, PAOD detection, and PAD classification with original images was observed. A similar observation was found for PAD detection, PAOD detection, and PAD classification with background-removal or cropped images. GP group had a lower performance for PAD and PAOD detection than did the three models with original images, while a similar performance for PAD detection was observed in PRS group and the 3 models.</div></div><div><h3>Conclusions</h3><div>We proposed a promising framework using CNN-based deep learning based on objective ankle-brachial index (ABI) values and image preprocessing to characterize PAD detection, PAOD detection, and PAD classification for lower extremity wounds, which provides an easily implemented and objective and reliable computational tool for physicians to automatically identify and classify PAD.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"263 ","pages":"Article 108654"},"PeriodicalIF":4.9,"publicationDate":"2025-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143445395","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":"Measures of spectral similarities for the detection of eye alignment during retinal birefringence scanning","authors":"Boris I. Gramatikov","doi":"10.1016/j.cmpb.2025.108650","DOIUrl":"10.1016/j.cmpb.2025.108650","url":null,"abstract":"<div><h3>Objective</h3><div>Retinal birefringence scanning is a well-established method for detecting central fixation. Using this technique, binocular eye alignment is confirmed when both eyes simultaneously fixate on a small target. Central fixation is identified when the spectral power of the scanning signal returned from the retina exceeds a certain threshold at a characteristic frequency, or a combination of frequencies. Traditionally, this assessment is performed separately for each eye, with binocular fixation declared when both pass the same threshold. However, factors such as hardware asymmetries, pupil diameter variability, retinal reflectivity differences, or suboptimal eye positioning within the device's exit pupil can introduce inaccuracies in threshold-based decision-making. This pilot study explores cross-spectral methods to mitigate amplitude imbalances and improve reliability.</div></div><div><h3>Methods</h3><div>This research examines spectral similarities between the signals from both eyes, to establish a more robust identification of eye alignment, independent of amplitude asymmetry. Two primary techniques are proposed and tested: magnitude-squared coherence and the spectral correlation coefficient, both of which quantify spectral linkage between the eyes.</div></div><div><h3>Results</h3><div>Magnitude-squared coherence reliably identifies eye alignment even in systems with significant signal imbalances, providing a continuous trace from which an alignment threshold can easily be determined. The spectral correlation coefficient, while computationally faster, has a limited time resolution. Additionally, spectral traces can be re-balanced using a linear fit, enhancing visualization. An algorithm for detecting the misaligned eye is also introduced, with potential clinical relevance pending validation.</div></div><div><h3>Conclusions and significance</h3><div>The proposed spectral-domain techniques offer reliable measures of signal similarity for detecting eye alignment. These findings have the potential to significantly enhance the precision of decision-making in ophthalmic diagnostic devices utilizing retinal birefringence scanning. Of particular importance is their application in pediatric vision screeners, which play a crucial role in detecting strabismus (misaligned eyes) and amblyopia (\"lazy eye\").</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"261 ","pages":"Article 108650"},"PeriodicalIF":4.9,"publicationDate":"2025-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143372397","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}
Filipe Cerqueira , Marta Campos Ferreira , Maria Joana Campos , Carla Silvia Fernandes
{"title":"PocketOnco®: Prototyping a mobile app for health literacy and self-management of oncological diseases","authors":"Filipe Cerqueira , Marta Campos Ferreira , Maria Joana Campos , Carla Silvia Fernandes","doi":"10.1016/j.cmpb.2025.108649","DOIUrl":"10.1016/j.cmpb.2025.108649","url":null,"abstract":"<div><h3>Background</h3><div>The study aims to present and explain the development stages of a mobile app designed to improve health literacy for self-management of oncological diseases. Through the integration of gamification, the app aims to enhance patient engagement and education in an interactive manner.</div></div><div><h3>Methods</h3><div>The methodology of Design Science in Information Systems and Software Engineering was employed, which included stages of needs identification, requirements definition, prototyping, and iterative validation of the developed artifact. A total of 132 participants, consisting of patients and healthcare professionals, were involved in the development of the PocketOnco application. The subsequent implementation of the App, PocketOnco, involved usability testing, System Usability Scale assessment, and the collection of qualitative feedback.</div></div><div><h3>Results</h3><div>The usability testing analysis revealed excellent acceptance of PocketOnco, with the gamified elements such as quizzes and reward systems being particularly appreciated for their ability to consistently engage and motivate users<strong>.</strong></div></div><div><h3>Conclusion</h3><div>The various stages in the development of this resource ensure the quality of its purpose. The application proved to be a viable and attractive solution for both patients and healthcare professionals, suggesting a promising path for future digital interventions in the field of oncology.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"261 ","pages":"Article 108649"},"PeriodicalIF":4.9,"publicationDate":"2025-02-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143349314","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":"Enhancing thrombosis prevention in medical devices: The role of turbulence in washout performance using FDA benchmark nozzle model","authors":"Peng Fang , Peng Wu , Haiquan Feng , Haimei Huang","doi":"10.1016/j.cmpb.2025.108647","DOIUrl":"10.1016/j.cmpb.2025.108647","url":null,"abstract":"<div><h3>Background and objectives</h3><div>Thrombosis presents a significant and potentially lethal risk in medical devices. Turbulence has been associated with increased thrombosis risk, primarily due to heightened shear stress and resultant blood damage. However, it can be inferred that turbulence might also enhance washout performance through efficient transport and mixing, thereby mitigating thrombosis. This study explores the underappreciated role of turbulence.</div></div><div><h3>Methods</h3><div>The FDA benchmark nozzle model was used as a representative framework for medical devices. To elucidate the isolated role of turbulence on washout performance, comparative simulations were conducted at Reynolds numbers of 500 and 6500 using Large Eddy Simulation (LES) and Menter's Shear Stress Transport (SST) k-ω turbulence models. Washout performance, a critical indicator in thrombosis, is evaluated by a passive scalar transport model.</div></div><div><h3>Results</h3><div>The validation results align well with published data, confirming the reliability of the simulations. Reynolds numbers and turbulence models play a crucial role in the washout performance. Turbulence improves volume washout by disrupting flow recirculation zones and enhancing the mixing of old and new blood. Furthermore, turbulence aids in surface washout by altering flow patterns in the near-wall region and increasing wall shear stress.</div></div><div><h3>Conclusion and significance</h3><div>The improved washout and dynamic environment facilitated by turbulence potentially minimize platelet adhesion and aggregation, which ultimately benefits the anti-thrombotic properties of medical devices. This research offers a novel perspective on the role of turbulence in thrombosis, extending beyond its traditionally recognized detrimental effects, and provides valuable insights into the design of specific flow patterns in achieving optimal washout performance in medical device applications. Further research is warranted to explore how to effectively leverage the washout-enhancing effects of turbulence while minimizing its potential adverse impacts.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"261 ","pages":"Article 108647"},"PeriodicalIF":4.9,"publicationDate":"2025-02-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143372398","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}
Haoran Dou , Yuhao Huang , Yunzhi Huang , Xin Yang , Chaojiong Zhen , Yuanji Zhang , Yi Xiong , Weijun Huang , Dong Ni
{"title":"Standard plane localization using denoising diffusion model with multi-scale guidance","authors":"Haoran Dou , Yuhao Huang , Yunzhi Huang , Xin Yang , Chaojiong Zhen , Yuanji Zhang , Yi Xiong , Weijun Huang , Dong Ni","doi":"10.1016/j.cmpb.2025.108619","DOIUrl":"10.1016/j.cmpb.2025.108619","url":null,"abstract":"<div><h3>Background and Objective:</h3><div>Standard planes (SPs) acquisition is a fundamental yet crucial step in routine ultrasound (US) examinations. Compared to the 2D US, 3D US offers the advantage of capturing multiple SPs in a single scan, and visualizing particular SPs (e.g., the coronal plane of the uterus). However, SPs localization in 3D US is challenging due to the vast 3D search space, anatomical variability, and poor image quality.</div></div><div><h3>Methods:</h3><div>In this study, we present a probabilistic method based on the conditional denoising diffusion model for SPs localization in 3D US. Specifically, we construct multi-scale guidance to provide the model with both global and local context. We improve the model’s angular sensitivity by modifying the tangent-based plane representation with the spherical coordinates. We also reveal the potential in simultaneously localizing SPs and detecting their abnormality without introducing extra parameters.</div></div><div><h3>Results:</h3><div>Extensive validations were performed on a large in-house dataset containing 837 patients across two organs with four SPs. The proposed method achieved average errors of less than <span><math><mrow><mn>10</mn><mo>°</mo></mrow></math></span> and 1 mm in terms of the angle and distance on the four investigated SPs. Furthermore, it can obtain over 90% accuracy for detecting anomalies by simply thresholding the quantified uncertainty.</div></div><div><h3>Conclusions:</h3><div>The results show that our proposed method significantly outperformed the current state-of-the-art approaches regarding spatial and content metrics across four SPs in two organs, indicating its superiority and generalizability. Meanwhile, the investigated anomaly detection of our method demonstrates its potential in applying clinical practice.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"261 ","pages":"Article 108619"},"PeriodicalIF":4.9,"publicationDate":"2025-02-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143294121","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":"Antimicrobial resistance recommendations via electronic health records with graph representation and patient population modeling","authors":"Pei Gao , Zheng Chen , Xin Liu , Peng Chen , Yasuko Matsubara , Yasushi Sakurai","doi":"10.1016/j.cmpb.2025.108616","DOIUrl":"10.1016/j.cmpb.2025.108616","url":null,"abstract":"<div><h3>Background:</h3><div>Antimicrobial resistance (AMR), which refers to the ability of pathogenic bacteria to withstand the effects of antibiotics, is a critical global health issue. Traditional methods for identifying AMRs in clinical settings rely on in-lab testing, which hampers timely medical decision-making. Moreover, there is a notable delay in updating empirical treatment guidelines in response to the rapid evolution of pathogens. Recent advances in AMR research have illuminated the potential of machine learning-based patient information analysis using electronic health records (EHRs).</div></div><div><h3>Methods:</h3><div>Against this backdrop, our study introduces a novel deep learning framework designed to leverage EHR data for generating AMR recommendations. This framework is anchored in three critical innovations. Firstly, we employ a deep graph neural network to model the correlations between various medical events, using structural information to enhance the representation of binary medical events. Secondly, in acknowledgment of the commonalities in pathogen evolution among populations, we incorporate population-level observation by modeling patient graphical structures. This strategy also addresses the issue of imbalance in rare AMR labels. Finally, we adopt a multi-task learning strategy, enabling simultaneous recommendations on multiple AMRs. Extensive experimental evaluations on a large dataset of over 110,000 patients with urinary tract infections validate the superiority of our approach.</div></div><div><h3>Results:</h3><div>It achieves notable improvements in areas under receiver operating characteristic curves (AUROCs) for four distinct AMR labels, with increments of 0.04, 0.02, 0.06, and 0.10 surpassing the baselines.</div></div><div><h3>Conclusions:</h3><div>Further medical analysis underscores the efficacy of our approach, demonstrating the potential of EHR-based systems in AMR recommendation.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"261 ","pages":"Article 108616"},"PeriodicalIF":4.9,"publicationDate":"2025-02-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143294126","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}
Frederic Jonske , Moon Kim , Enrico Nasca , Janis Evers , Johannes Haubold , René Hosch , Felix Nensa , Michael Kamp , Constantin Seibold , Jan Egger , Jens Kleesiek
{"title":"Why does my medical AI look at pictures of birds? Exploring the efficacy of transfer learning across domain boundaries","authors":"Frederic Jonske , Moon Kim , Enrico Nasca , Janis Evers , Johannes Haubold , René Hosch , Felix Nensa , Michael Kamp , Constantin Seibold , Jan Egger , Jens Kleesiek","doi":"10.1016/j.cmpb.2025.108634","DOIUrl":"10.1016/j.cmpb.2025.108634","url":null,"abstract":"<div><h3>Purpose</h3><div>In medical deep learning, models not trained from scratch are typically fine-tuned based on ImageNet-pretrained models. We posit that pretraining on data from the domain of the downstream task should almost always be preferable.</div></div><div><h3>Materials and methods</h3><div>We leverage RadNet-12M and RadNet-1.28M, datasets containing >12 million/1.28 million acquired CT image slices from 90,663 individual scans, and explore the efficacy of self-supervised, contrastive pretraining on the medical and natural image domains. We compare the respective performance gains for five downstream tasks. For each experiment, we report accuracy, AUC, or DICE score and uncertainty estimations based on four separate runs. We quantify significance using Welch's <em>t</em>-test. Finally, we perform feature space analysis to characterize the nature of the observed performance gains.</div></div><div><h3>Results</h3><div>We observe that intra-domain transfer (RadNet pretraining and CT-based tasks) compares favorably to cross-domain transfer (ImageNet pretraining and CT-based tasks), generally achieving comparable or improved performance – Δ = +0.44% (<em>p</em> = 0.541) when fine-tuned on RadNet-1.28M, Δ = +2.07% (<em>p</em> = 0.025) when linearly evaluating on</div><div>RadNet-1.28M, and Δ = +1.63% (<em>p</em> = 0.114) when fine-tuning on 1 % of RadNet-1.28M data. This intra-domain advantage extends to LiTS 2017, another CT-based dataset, but not to other medical imaging modalities. A corresponding intra-domain advantage was also observed for natural images. Outside the CT image domain, ImageNet-pretrained models generalized better than RadNet-pretrained models.</div><div>We further demonstrate that pretraining on medical images yields domain-specific features that are preserved during fine-tuning, and which correspond to macroscopic image properties and structures.</div></div><div><h3>Conclusion</h3><div>We conclude that intra-domain pretraining generally outperforms cross-domain pretraining, but that very narrow domain definitions apply. Put simply, pretraining on CT images instead of natural images yields an advantage when fine-tuning on CT images, and only on CT images. We further conclude that ImageNet pretraining remains a strong baseline, as well as the best choice for pretraining if only insufficient data from the target domain is available. Finally, we publish our pretrained models and pretraining guidelines as a baseline for future research.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"261 ","pages":"Article 108634"},"PeriodicalIF":4.9,"publicationDate":"2025-01-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143294122","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}
Yining Zhang , Zhongze Cao , Xiran Cao , Yue Che , Xuelan Zhang , Mingyao Luo , Chang Shu
{"title":"Hemodynamics of different surgical subclavian revascularization morphologies for thoracic endovascular aortic repair","authors":"Yining Zhang , Zhongze Cao , Xiran Cao , Yue Che , Xuelan Zhang , Mingyao Luo , Chang Shu","doi":"10.1016/j.cmpb.2025.108632","DOIUrl":"10.1016/j.cmpb.2025.108632","url":null,"abstract":"<div><h3>Background and Objective</h3><div>Carotid-subclavian bypass (CSB) and subclavian-carotid transposition (SCT) are mainstream surgical left subclavian artery (LSA) revascularization methods. However, surgical selection of CSB and SCT morphological configurations mainly depends on surgeons’ experience, lacking objective data basis.</div></div><div><h3>Methods</h3><div>Geometries with 28 configurations, including length, diameter, angle, and anastomotic direction for prosthetic conduit and transposed LSA, were constructed. Numerical simulations were performed to evaluate CSB and SCT outcomes by hemodynamic parameters such as pressure drop, flow rate, energy loss and wall shear stress related indicators.</div></div><div><h3>Results</h3><div>After CSB, enlarging prosthetic conduit diameter (6 to 10 mm) increases flow rate by 36.64 %, suggesting larger diameter enhances LSA patency. However, when diameter exceeds 9 mm, the relative residence time rises by 35.29 %, demonstrating oversized diameter increases the risk of thrombosis. Compared to 5 mm, prosthetic conduit at 15 mm displays a 7.80 % flow rate reduction, indicating longer conduit causes greater flow resistance. For varying angles, prosthetic conduit perpendicular to left common carotid artery (LCCA) shows the least energy loss. Conduit tilted downward from the vertical position shows higher flow rate than the upward during systole (210.35 vs. 106.34 ml/min). However, 10 % blood flow in downward conduit reflows cyclically during diastole, resulting in the reduced cycle-averaged flow rate of downward conduit compared to that of the upward (53.21 vs. 58.42 ml/min). After SCT, configurations with smaller angles between LCCA and LSA show better hemodynamic performance, with a maximum flow rate variation of 30.34 % in LSA from 50° to 110°.</div></div><div><h3>Conclusions</h3><div>Configurations with moderately smaller diameter, reduced length of prosthetic conduit and aligned anastomosis towards LCCA blood flow result in better LSA revascularization outcomes. The findings are supportive for optimizing CSB and SCT configurations.</div></div>","PeriodicalId":10624,"journal":{"name":"Computer methods and programs in biomedicine","volume":"261 ","pages":"Article 108632"},"PeriodicalIF":4.9,"publicationDate":"2025-01-31","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143104379","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}