Computers in biology and medicine最新文献

{"title":"A novel methodology for calibration and experimental validation of cervical spine biomechanics models using ANFIS and genetic algorithm","authors":"Ali Cherif Messellek ,&nbsp;Khalil Chenaifi ,&nbsp;Mohamed Medaouar ,&nbsp;Abdelghani May ,&nbsp;Mohand Ould Ouali ,&nbsp;Samir Khatir ,&nbsp;Thanh Cuong-Le","doi":"10.1016/j.compbiomed.2025.111105","DOIUrl":"10.1016/j.compbiomed.2025.111105","url":null,"abstract":"<div><div>The cervical spine is an essential structure in human physiology but is vulnerable to diseases, such as lesions, disc herniation, and vertebral fractures. Finite element (FE) modeling represents a powerful approach for predicting the biomechanics of the cervical spine under various loading conditions. Conventional methods usually do not take into account the consistency of the material properties, which potentially restricts their capability to realistically represent biomechanical behavior. This study presents a novel calibration methodology aimed at enhancing the predictive biofidelity of FE models for cervical spine biomechanics. The proposed approach systematically identifies optimal material properties to better replicate in vitro biomechanical responses. To achieve this, the cervical spine geometry was reconstructed from computed tomography (CT) scans and validated against cadaveric morphometric data to ensure anatomical accuracy. Mechanical properties of both hard and soft tissues were collected based on a comprehensive review of the literature. A dataset of biomechanical responses was generated through FE simulations using a range of material properties. The calibration process integrated an adaptive neuro-fuzzy inference system (ANFIS) framework with genetic algorithms, followed by a rigorous validation step against experimental benchmarks to ensure precise replication of in vitro test outcomes. The results show that this calibrated FE model significantly improves cervical spine biomechanics predictions, accurately matching intervertebral disc mechanics and ligament behavior. This research provides a robust framework, integrating numerical modeling with experimental studies, guiding future biomechanical research to potentially improve clinical and surgical outcomes. Some of the prominent applications are injury analysis, degenerative disease modeling, and the prediction of spinal deformity.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"197 ","pages":"Article 111105"},"PeriodicalIF":6.3,"publicationDate":"2025-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145091390","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":"Markov chain-based computational model to assess user skills in sequential motor imagery tasks","authors":"Cristian David Guerrero-Mendez ,&nbsp;Hamilton Rivera-Flor ,&nbsp;Denis Delisle-Rodriguez ,&nbsp;Leonardo Abdala Elias ,&nbsp;Teodiano Freire Bastos-Filho","doi":"10.1016/j.compbiomed.2025.111048","DOIUrl":"10.1016/j.compbiomed.2025.111048","url":null,"abstract":"<div><div>Researchers face challenges in accurately measuring individual skills to generate discernible patterns during motor imagery (MI). Despite this, most studies focus on simple tasks, limiting knowledge of the effects on more complex sequential motor activities. This study proposes and evaluates a computational method based on state clustering and Markov chains to assess user skills during sequential MI tasks, such as object manipulation. The method is evaluated using two Markov-derived metrics—<em>taskDistinct</em> and <em>relativeTaskInconsistency</em>—which capture, respectively, inter-task pattern separability and consistency of cortical states. A dataset from 30 healthy participants who had to sequentially imagine the manipulation of an object was used, where the cup’s locations were varied across four positions (right, left, up, and down). The analyses performed determine whether the extracted cortical pattern states and their transitions, using the proposed method, reflected the number and structure of sequential actions involved in the imagined task, and also whether the Markov-derived metrics correlated with conventional classification metrics. Results showed that the number of states during training matched the number of imagined actions, while during testing, slightly fewer states were recognized. However, the transitions preserved the expected ascending order of the task. Correlations between Markov-based and conventional metrics varied by task, highlighting the task-dependent nature of the associations, where significant and non-significant correlations were observed. Notably, the right-cup MI exhibited predominantly significant correlations. In conclusion, the proposed method successfully captures features of sequential MI performance and provides complementary information about user skills, beyond what classification metrics alone reveals.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"197 ","pages":"Article 111048"},"PeriodicalIF":6.3,"publicationDate":"2025-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145091359","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":"Building hybrid models of neuromodulation from automatic segmentation of peripheral nerve histological sections","authors":"Claudio Verardo ,&nbsp;Alice Giannotti ,&nbsp;Clement Albert ,&nbsp;Giovanni Faoro ,&nbsp;Chléa Schiff ,&nbsp;Justine Bourgeot ,&nbsp;Giulia Lazzarini ,&nbsp;Andrea Pirone ,&nbsp;Vincenzo Miragliotta ,&nbsp;Sara Moccia ,&nbsp;Silvestro Micera ,&nbsp;Simone Romeni","doi":"10.1016/j.compbiomed.2025.111072","DOIUrl":"10.1016/j.compbiomed.2025.111072","url":null,"abstract":"<div><div>Electrical stimulation of peripheral nerves offers a way to restore sensory-motor functions and treat drug-resistant conditions affecting internal organs. Understanding the fascicular organization of the implanted nerves is essential for enhancing the selective neuromodulation of the targeted bodily functions. In fact, this knowledge can inform the development of computational models that can be used to optimize electrode design and stimulation protocols. Traditionally, peripheral nerve topographies are segmented manually to highlight fascicle contours, resulting in a labor-intensive and error-prone process. In this study, we present a UNet-based deep neural network for automatic segmentation of fascicles from nerve histological sections, trained on original data from different nerves and stained with different techniques. The model leverages a pretrained encoder, reducing the need for extensive training datasets and allowing us to generalize to nerve types and histological stains previously unseen during training. The quality of the resulting segmentation has been evaluated using both the Dice coefficient and domain-specific metrics tailored to assess the quality of the reconstructed fascicle topography. Furthermore, we employed automatically segmented nerve sections to build computational models of peripheral nerve stimulation and assess the impact of segmentation on the accuracy of fascicle-wise recruitment predictions. Our results highlight that automated segmentation can reliably inform the modeling of neuromodulation applications, with minimal error in predicting recruitment thresholds. This approach paves the way for harnessing the large quantities of histological data that can be extracted from cadaveric nerve samples for use in computational models of neural interfaces, potentially advancing the design of next generation neuroprosthetic and bioelectronic medicine applications.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"197 ","pages":"Article 111072"},"PeriodicalIF":6.3,"publicationDate":"2025-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145091318","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":"Bridging the quality gap: Robust colon wall segmentation in noisy transabdominal ultrasound","authors":"Lucas Gago ,&nbsp;Miguel A. Fernández González ,&nbsp;Justin Engelmann ,&nbsp;Beatriz Remeseiro ,&nbsp;Laura Igual","doi":"10.1016/j.compbiomed.2025.111077","DOIUrl":"10.1016/j.compbiomed.2025.111077","url":null,"abstract":"<div><div>Colon wall segmentation in transabdominal ultrasound is challenging due to variations in image quality, speckle noise, and ambiguous boundaries. Existing methods struggle with low-quality images due to their inability to adapt to varying noise levels, poor boundary definition, and reduced contrast in ultrasound imaging, resulting in inconsistent segmentation performance. We present a novel quality-aware segmentation framework that simultaneously predicts image quality and adapts the segmentation process accordingly. Our approach uses a U-Net architecture with a ConvNeXt encoder backbone, enhanced with a parallel quality prediction branch that serves as a regularization mechanism. Our model learns robust features by explicitly modeling image quality during training. We evaluate our method on the C-TRUS dataset and demonstrate superior performance compared to state-of-the-art approaches, particularly on challenging low-quality images. Our method achieves Dice scores of 0.7780, 0.7025, and 0.5970 for high, medium, and low-quality images, respectively. The proposed quality-aware segmentation framework represents a significant step toward clinically viable automated colon wall segmentation systems.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"197 ","pages":"Article 111077"},"PeriodicalIF":6.3,"publicationDate":"2025-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145091401","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":"Tsallis statistics enhanced logistic regression for gene expression classification","authors":"Baiyang Zhang ,&nbsp;Shunjie Chen ,&nbsp;Keming Shen ,&nbsp;Yang Wu ,&nbsp;Pei Wang ,&nbsp;Liugen Xue","doi":"10.1016/j.compbiomed.2025.111062","DOIUrl":"10.1016/j.compbiomed.2025.111062","url":null,"abstract":"<div><div>Two kinds of Tsallis statistics-enhanced sigmoid functions are introduced based on <span><math><mi>q</mi></math></span>-deformed exponents for <span><math><mrow><mi>q</mi><mo>&lt;</mo><mn>1</mn></mrow></math></span> that are free of empirically chosen cutoffs. These generalizations of the classical sigmoid enables a more flexible and robust fitting method in the context of classification problems, particularly when dealing with complex, non-linear dependencies in data. The <span><math><mi>q</mi></math></span>-deformed classifiers are applied to four cancer datasets, demonstrating its robustness, noise resistance, and stability. In the simulated experiments, the improved algorithm outperforms traditional methods such as Logistic Regression, SVM, and Random Forest, with significantly smaller standard deviation. On real cancer datasets, Tsallis enhanced method achieves substantial improvements, particularly outperforming Logistic Regression with traditional sigmoid function with breast cancer data. These results demonstrate the exceptional robustness, noise resistance and stability of Tsallis statistics-enhanced method, making it a reliable solution for complex and noisy data environments.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"197 ","pages":"Article 111062"},"PeriodicalIF":6.3,"publicationDate":"2025-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145085356","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":"Predicting dementia through audio: Ensemble and deep learning approaches using acoustic features","authors":"Priyanka G. ,&nbsp;Amshakala K.","doi":"10.1016/j.compbiomed.2025.111078","DOIUrl":"10.1016/j.compbiomed.2025.111078","url":null,"abstract":"<div><div>A deterioration in cognitive function beyond what one might expect from normal aging characterizes the symptoms of dementia. It predominantly marks older adults, although it is not a normal part of aging. Dementia encompasses a range of symptoms that can include memory loss, impaired reasoning, personality changes, and difficulties with daily activities. One of the major difficulties that elderly people with dementia tend to face is communicating with other people to meet their daily needs. Diagnosing dementia involves a comprehensive evaluation of an individual's cognitive function, medical history, and other relevant factors. In this work, audio recordings of patients are used for diagnosing dementia at earlier stages. To do this, we take sound characteristics from the audio recordings, such as pitch, variations in pitch, loudness changes, how quickly the voice starts, and specific sound patterns. We then selected the best acoustic features using statistical methods to train ensemble models such as Random Forest, AdaBoost, XGBoost, and Gradient Boost. In addition to ensemble learning models, certain deep learning models like BiLSTM, LSTM, and CNN-LSTM are also trained with these features. The features selected for training include spectral centroid, MFCC, and fundamental frequency (F0). Further, both the ensemble learning models and the deep learning models underwent random search for hyperparameter tuning, along with regularization and cross-validation, to enhance their performance. It was observed that the gradient boost model was found to perform well with an accuracy of 90.5 % in diagnosing dementia from audio data when trained with spectral centroid, MFCC, and fundamental frequency (F0). Furthermore, the study explores the underlying factors that may lead ensemble models to achieve superior performance over deep learning models in specific cases, even though deep learning models are typically considered more effective for large-scale datasets.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"197 ","pages":"Article 111078"},"PeriodicalIF":6.3,"publicationDate":"2025-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145085277","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":"Characterizing resting-state EEG oscillatory and aperiodic activity in neurodegenerative diseases: A multicentric study","authors":"Alberto Jaramillo-Jimenez ,&nbsp;Yorguin-Jose Mantilla-Ramos ,&nbsp;Diego A. Tovar-Rios ,&nbsp;Francisco Lopera ,&nbsp;David Aguillón ,&nbsp;John Fredy Ochoa-Gomez ,&nbsp;Claire Paquet ,&nbsp;Sinead Gaubert ,&nbsp;Matteo Pardini ,&nbsp;Dario Arnaldi ,&nbsp;John-Paul Taylor ,&nbsp;Tormod Fladby ,&nbsp;Kolbjørn Brønnick ,&nbsp;Dag Aarsland ,&nbsp;Laura Bonanni ,&nbsp;E-DLB Consortium","doi":"10.1016/j.compbiomed.2025.111080","DOIUrl":"10.1016/j.compbiomed.2025.111080","url":null,"abstract":"<div><h3>Background</h3><div>Abnormalities in resting-state electroencephalogram (rsEEG) posterior alpha rhythm are promising biomarkers of neurodegenerative diseases (NDDs), often assessed via spectral analysis, ignoring the signal's non-rhythmic (aperiodic) component. Evidence assessing aperiodic and oscillatory rsEEG abnormalities across NDDs is scarce and often underpowered. Multicenter studies could tackle these limitations, but data pooling might introduce site-related rsEEG differences (batch effects). This study aims to characterize rsEEG oscillatory and aperiodic patterns across NDDs, minimizing potential batch effects.</div></div><div><h3>Methods</h3><div>RsEEGs (n = 639; 11 sites) were automatically preprocessed. Signals comprised healthy controls (HC = 153), Lewy Body Dementias (LBD = 95), Parkinson's Disease (PD = 71), Alzheimer's Disease (AD = 186), Frontotemporal Dementia (FTD = 23), Mild Cognitive Impairment (MCI) in positive Lewy Bodies pathology or PD (MCI-LBD = 34), and MCI in positive AD pathology (MCI-AD = 77). Power spectrum batch effects were harmonized using reComBat (age, sex, and diagnosis-adjusted). Harmonization was evaluated with functional and mass-univariate ANOVAs. Oscillatory and aperiodic parameters were extracted from the batch-harmonized power spectrum. NDDs-related differences were estimated with functional and mass-univariate tests, bootstrapped pairwise comparisons, and logistic regressions.</div></div><div><h3>Results</h3><div>Statistical testing showed reduced batch effects after harmonization. Significantly steeper aperiodic parameters and lower oscillatory center frequency were observed in LBD compared to other NDDs. The oscillatory extended alpha power was lower in AD comparisons (except AD vs. LBD).</div></div><div><h3>Conclusions</h3><div>Batch effects in the rsEEG power spectrum can be mitigated with harmonization. Oscillatory alpha power reduction may better reflect AD abnormalities, whereas pronounced oscillatory frequency slowing and greater aperiodic activity characterize LBD.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"197 ","pages":"Article 111080"},"PeriodicalIF":6.3,"publicationDate":"2025-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145085306","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":"Classification of peripheral pulmonary lesions in Endobronchial ultrasonography image using a multi-branch framework and voting ensemble","authors":"Huitao Wang ,&nbsp;Takahiro Nakajima ,&nbsp;Kohei Shikano ,&nbsp;Yukihiro Nomura ,&nbsp;Toshiya Nakaguchi","doi":"10.1016/j.compbiomed.2025.111064","DOIUrl":"10.1016/j.compbiomed.2025.111064","url":null,"abstract":"<div><h3>Background and Objective:</h3><div>Lung cancer stands as a significant contributor to cancer-related fatalities worldwide. Endobronchial ultrasonography plays a crucial role in the early diagnosis of lung cancer. In this study, our objective is to formulate a deep learning-based Computer-aided Diagnosis (CAD) system for lung cancer, aiming to assist medical professionals in achieving more precise and efficient diagnoses.</div></div><div><h3>Method:</h3><div>In this research, acknowledging the pronounced issue of extreme data imbalance, we propose a multi-branch framework. Additionally, to enhance the performance of the CAD system further, we employ a majority voting mechanism, integrating multiple branches to generate the final output. The design of this multi-branch framework aims to better adapt to the distribution differences among various categories, thereby augmenting the model’s capability to recognize minority classes. Furthermore, we explored a coordinate system transformation approach, wherein the original Endobronchial Ultrasonography (EBUS) images are converted from polar coordinates to Cartesian coordinates. Such a transformation may contribute to reducing the complexity of image processing, providing deep learning models with clearer and more consistent inputs, thereby augmenting the model’s ability to extract features related to lung cancer.</div></div><div><h3>Results:</h3><div>The proposed multi-branch CAD diagnostic system, utilizing EBUS images transformed through coordinate system conversion, has demonstrated good performance. This method achieved a level of 0.80 in terms of Area Under the Curve (AUC), with an accuracy of 0.78, F1 score of 0.80, positive predictive value of 0.77, negative predictive value of 0.83, sensitivity of 0.85, and specificity of 0.72.</div></div><div><h3>Conclusion:</h3><div>The utilization of a multi-branch framework and ensemble learning proves to be more effective in addressing data imbalance issues. Furthermore, image transformation based on coordinate systems contributes to optimizing the model’s understanding of image structures, which can further enhance performance.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"197 ","pages":"Article 111064"},"PeriodicalIF":6.3,"publicationDate":"2025-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145085372","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":"Emotion recognition in Virtual Reality using sensor fusion with eye tracking","authors":"Meral Kuyucu ,&nbsp;Mehmet Ali Sarikaya ,&nbsp;Tülay Karakaş ,&nbsp;Dilek Yıldız Özkan ,&nbsp;Yüksel Demir ,&nbsp;Ömer Bilen ,&nbsp;Gökhan Ince","doi":"10.1016/j.compbiomed.2025.111070","DOIUrl":"10.1016/j.compbiomed.2025.111070","url":null,"abstract":"<div><div>Emotion recognition is an emerging field with applications in healthcare, education, and entertainment. This study integrates Virtual Reality (VR) with multi-sensor fusion to enhance emotion recognition. The research comprises two phases: data collection and analysis/evaluation. Ninety-five participants were exposed to curated audiovisual stimuli designed to elicit a wide range of emotions within an immersive VR environment. VR was chosen for its ability to provide controlled conditions and overcome the limitations of current mobile sensor technologies. Physiological data streams from various sensors were integrated for comprehensive emotional analysis. ElectroEncephaloGraphy (EEG) data revealed brain activity linked to emotional states, while eye tracking data provided insights into gaze direction, pupil dilation, and eye movement—factors correlated with cognitive and emotional processes. Peripheral signals, including heart rate variability, ElectroDermal Activity (EDA), and body temperature, were captured via wearable sensors to enrich the dataset. Machine learning models, such as XGBoost, CatBoost, Multilayer Perceptron, Gradient Boosting, and LightGBM, were employed to predict participants’ emotional states. Evaluation metrics, including accuracy, precision, recall, and F1 scores, demonstrated the robustness and precision of the proposed VR-based multi-sensor fusion approach. This research presents a novel approach to emotion recognition, bridging gaps in traditional methods by integrating VR, multi-sensor fusion, and machine learning.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"197 ","pages":"Article 111070"},"PeriodicalIF":6.3,"publicationDate":"2025-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145085374","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":"Robust and explainable framework to address data scarcity in diagnostic imaging","authors":"Zehui Zhao ,&nbsp;Laith Alzubaidi ,&nbsp;Jinglan Zhang ,&nbsp;Ye Duan ,&nbsp;Usman Naseem ,&nbsp;Yuantong Gu","doi":"10.1016/j.compbiomed.2025.111052","DOIUrl":"10.1016/j.compbiomed.2025.111052","url":null,"abstract":"<div><div>Deep learning has significantly advanced automatic medical diagnostics, releasing human resources from clinical pressure, yet the persistent challenge of data scarcity in this area hampers its further improvements and applications. To address this gap, we introduce a novel ensemble framework called ‘Efficient Transfer and Self-supervised Learning based Ensemble Framework’ (ETSEF). ETSEF leverages features from multiple pre-trained deep learning models to efficiently learn powerful representations from a limited number of data samples. To the best of our knowledge, ETSEF is the first strategy that combines two pre-training methodologies (Transfer Learning and Self-supervised Learning) with ensemble learning approaches. Various data enhancement techniques, including data augmentation, feature fusion, feature selection, and decision fusion, have also been deployed to maximise the efficiency and robustness of the ETSEF model. Five independent medical imaging tasks, including endoscopy, breast cancer detection, monkeypox detection, brain tumour detection, and glaucoma detection, were tested to demonstrate ETSEF’s effectiveness and robustness. Facing limited sample numbers and challenging medical tasks, ETSEF has demonstrated its effectiveness by improving diagnostic accuracy by up to 13.3% compared to strong ensemble baseline models and up to 14.4% compared with recent state-of-the-art methods. Moreover, we emphasise the robustness and trustworthiness of the ETSEF method through various vision-explainable artificial intelligence techniques, including Grad-CAM, SHAP, and t-SNE. Compared to large-scale deep learning models, ETSEF can be flexibly deployed and maintain superior performance for challenging medical imaging tasks, demonstrating potential for application in areas lacking training data. The code is available at <span><span>Github ETSEF</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":10578,"journal":{"name":"Computers in biology and medicine","volume":"197 ","pages":"Article 111052"},"PeriodicalIF":6.3,"publicationDate":"2025-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145085367","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}