Informatics in Medicine Unlocked最新文献

{"title":"DenseNet121 with ghost modules and coordinate attention for breast tumor classification in ultrasound images","authors":"Ameera Jaradat,&nbsp;Amal Alishwait,&nbsp;Majd Malkawi,&nbsp;Amer Mahmoud Sindiani,&nbsp;Hamad Yahia Abu Mhanna,&nbsp;Salem Alhatamleh,&nbsp;Qutaiba Mohammad,&nbsp;Malik H. Al-Soukhni,&nbsp;Mohammad Amin,&nbsp;Alaa Abd-alrazaq","doi":"10.1016/j.imu.2026.101822","DOIUrl":"10.1016/j.imu.2026.101822","url":null,"abstract":"<div><div>Breast cancer is one of the leading causes of cancer-related death among women. Early detection of breast tumors through ultrasound imaging can facilitate timely clinical intervention and improve patient outcomes. This study proposes a novel deep learning architecture, Bio-GhostV2-DenseNet, which uses DenseNet121 as the backbone feature extractor, Ghost modules for computationally efficient feature generation, and Coordinate Attention to capture long-range spatial dependencies while preserving positional information. This design enables the model to learn discriminative features while reducing redundant feature computation. The model was trained and tested on a real-world clinical dataset collected at King Abdullah University Hospital, Jordan University of Science and Technology, Jordan, to classify breast tumors as benign or malignant. The proposed model achieved an accuracy of 94.25%, a recall of 91.79%, and an F1-score of 94.11%, outperforming five established deep learning models (MobileNetV2, Xception, NASNetMobile, InceptionV3, and DenseNet121) in terms of accuracy and F1-score. On the original, non-augmented test set, it maintained an accuracy of 93.67% and a recall of 88.37%. When retrained and evaluated on the publicly available three-class Breast Ultrasound Images (BUSI) dataset, the model achieved an accuracy of 88.46%, indicating that the proposed architecture generalizes across datasets. The findings suggest that combining Coordinate Attention with Ghost-based feature generation may improve breast ultrasound classification performance.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"66 ","pages":"Article 101822"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149155969","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":"Artificial intelligence in Parkinson's disease: A critical review of multimodal biomarkers, validation, and clinical translation","authors":"Kagne Raveena Rajendra,&nbsp;Jerry Gao,&nbsp;Roopa Gangadhar Devihosur,&nbsp;Mohini Kathrotiya","doi":"10.1016/j.imu.2026.101824","DOIUrl":"10.1016/j.imu.2026.101824","url":null,"abstract":"<div><div>Parkinson's disease (PD) is a heterogeneous neurodegenerative disorder whose early diagnosis and longitudinal assessment remain difficult because motor and non-motor manifestations vary across patients and clinically evident symptoms often emerge after substantial dopaminergic degeneration. Artificial intelligence (AI) can integrate heterogeneous biomarkers, but the reliability of reported performance depends strongly on dataset composition, validation design, model interpretability, and external generalisability. This PRISMA-guided review qualitatively synthesised 74 eligible studies and this review therefore evaluates how machine learning (ML), deep learning (DL), multimodal learning, and explainable and privacy-preserving AI have been used for PD detection, severity assessment, progression modelling, digital-biomarker discovery, and AI-assisted drug discovery.</div><div>The review provides a dataset-centric assessment of more than 20 publicly available benchmark datasets spanning seven biomarker modalities, including speech, gait and wearable signals, handwriting, neuroimaging, clinical assessments, molecular and multi-omics data, and multimodal combinations. Across these areas, conventional ML remains useful for smaller structured datasets, whereas DL supports automated feature learning from high-dimensional signals and images; multimodal approaches may capture complementary disease information but require larger harmonised cohorts and more complex validation. The evidence is limited by small or imbalanced samples, inconsistent evaluation protocols, overfitting, scarce independent external validation, restricted interpretability, and demographic and acquisition-related dataset shift. Consequently, high within-dataset performance should not be interpreted as clinical readiness. Future progress requires standardised benchmarks, prospective multicentre validation, explainable multimodal models, longitudinal monitoring, and privacy-preserving data collaboration.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"66 ","pages":"Article 101824"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149201048","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":"EfficientNet-based detection transformer with explainable AI for tuberculosis bacilli detection and severity grading","authors":"Ashutosh Satapathy,&nbsp;L.M. Jenila Livingston,&nbsp;S. Rajkumar,&nbsp;Praneeth Vallabhaneni,&nbsp;Manisha Indugula,&nbsp;Jafflyn F. Leni","doi":"10.1016/j.imu.2026.101820","DOIUrl":"10.1016/j.imu.2026.101820","url":null,"abstract":"<div><div>TB continues to be a serious health issue in the world, with about 10 million people infected each year. Good quality tools that can help in making sound medical decisions are essential for improving the health status of patients. This paper explores an automatic TB bacilli detection and grading system based on artificial intelligence. The detection system is based on the Real-Time Detection Transformer (RT-DETR), allowing for accurate and effective bacilli identification at different scales. To improve model transparency and support clinician interpretability, Gradient-weighted Class Activation Mapping (Grad-CAM) was incorporated. The system incorporates an IUATLD-based grading scheme, endorsed by the World Health Organization (WHO), classifying the severity of infection into three levels: TB 1+, TB 2+, and TB 3+. The grades are based on bacilli counts seen on 300 slides or Fields of View (FoV), providing a standardized and objective measurement. The proposed RT-DETR with an EfficientNet-B0 backbone achieved a detection performance of 92.5% mAP@50, 84.8% recall, and 82.8% box precision on the TB Phone Camera dataset, indicating strong detection performance compared with previously reported YOLO-based approaches while preserving real-time inference performance. The proposed model achieved a severity grading accuracy of 94.45%. By integrating Explainable Artificial Intelligence (XAI) with established IUATLD grading criteria, the proposed framework provides an interpretable approach to automated TB detection and severity grading.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"66 ","pages":"Article 101820"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149155967","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":"CXRMate-2: Structured multimodal temporal embeddings and tractable reinforcement learning for clinically acceptable chest X-ray radiology report generation","authors":"Aaron Nicolson,&nbsp;Elizabeth J. Cooper,&nbsp;Hwan-Jin Yoon,&nbsp;Claire McCafferty,&nbsp;Ramya Krishnan,&nbsp;Michelle Craigie,&nbsp;Nivene Saad,&nbsp;Jason Dowling,&nbsp;Ian A. Scott,&nbsp;Bevan Koopman","doi":"10.1016/j.imu.2026.101816","DOIUrl":"10.1016/j.imu.2026.101816","url":null,"abstract":"<div><div>Chest X-ray (CXR) radiology report generation (RRG) models have made rapid progress on automated metrics, yet their clinical utility remains uncertain because few have undergone qualitative evaluation by radiologists. We present CXRMate-2, a state-of-the-art CXR RRG model designed to make reinforcement learning (RL) computationally tractable. In particular, it enables compute-intensive algorithms, such as group relative policy optimisation (GRPO), to achieve practical throughput with modest accelerator memory. This efficiency is achieved through structured multimodal temporal embeddings and compression of high-resolution visual features, which allow a large language model (LLM) decoder to be conditioned jointly on the visual, textual, and temporal context of a study and its prior examination. We also propose a reward function that improves semantic alignment between generated and radiologist reports. Across the MIMIC-CXR, CheXpert Plus, and ReXGradient datasets, CXRMate-2 achieves statistically significant improvements over strong benchmarks. On MIMIC-CXR, CXRMate-2, outperformed MedGemma 1.5 (4B) in RaTEScore (67.6 versus 64.1), GREEN (46.6 versus 41.9), and RadGraph-XL (40.3 versus 32.4).</div><div>To directly compare CXRMate-2 against radiologist reporting, we conducted a blinded, randomised qualitative retrospective evaluation. Three consultant radiologists compared generated and radiologist reports across 120 studies from the MIMIC-CXR test set. Generated reports were deemed acceptable — defined as preferred or judged equivalent to radiologist reports — in 45% of ratings. No statistically significant difference in acceptability from 50% was detected for seven of the eight analysed findings (all <span><math><mrow><mi>p</mi><mo>≥</mo><mn>0.211</mn></mrow></math></span>); pulmonary congestion was the exception (<span><math><mrow><mi>p</mi><mo>=</mo><mn>0.001</mn></mrow></math></span>). Recall and precision accounted for 57% and 27% of the reasons for preferring radiologist reports, respectively, whereas readability accounted for 50% of the reasons for preferring generated reports.</div><div>Together, these results define a clear pathway to clinically acceptable CXR RRG. The need to improve recall and detect subtle findings represents the primary remaining barrier to non-inferiority with radiologist reporting, positioning CXR RRG for prospective evaluation in assistive, radiologist-led workflows.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"66 ","pages":"Article 101816"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149155972","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":"GISR-Net: An image-to-scalar regressor for quality assessment of illumination correction in dermatological macro-photographs","authors":"Vipin Venugopal,&nbsp;Malaya Kumar Nath,&nbsp;Justin Joseph,&nbsp;V.R. Simi,&nbsp;Anoop B.N.","doi":"10.1016/j.imu.2026.101826","DOIUrl":"10.1016/j.imu.2026.101826","url":null,"abstract":"<div><div>Owing to their simplicity and flexibility, macroscopic photographs captured with ordinary digital cameras are commonly employed in dermatology as an alternative to conventional dermoscopy images. Macroscopic photographs often suffer from uneven background illumination, which degrades segmentation accuracy and diagnostic decisions in automated systems. Illumination correction is often used as a preprocessing step to address this issue. No-reference image quality assessment (NR-IQA) metrics are therefore necessary to evaluate the quality of macroscopic photographs at acquisition and after illumination correction. Such image quality assessment (IQA) metrics can also serve as a cost function for optimizing user-defined parameters of illumination correction algorithms. NR-IQA metrics whose quality ratings agree well with segmentation accuracy are not available. This paper proposes a task-based deep learning model, the GoogLeNet-based image-to-scalar regressor network (GISR-Net), for assessing the quality of illumination-corrected macroscopic photographs. GISR-Net is derived from GoogLeNet by modifying the final fully connected layer to produce a scalar output instead of class labels and by replacing the softmax and classification layers with a regression layer. Transfer learning was used to reduce the dataset size and training time required. To train GISR-Net, illumination-corrected images of different quality levels, generated by 16 state-of-the-art algorithms, are used as inputs, and the corresponding mean segmentation accuracy obtained with Otsu’s, minimum error threshold and Kapur’s maximum entropy segmentation is used as the target. The agreement of the proposed IQA metric with segmentation accuracy is evaluated using Pearson linear correlation coefficient (PLCC), Spearman’s rank-order correlation coefficient (SROCC), and Kendall’s rank-order correlation coefficient (KROCC). The proposed IQA metric (PLCC = 0.66, SROCC = 0.67, KROCC = 0.50) correlates more strongly with segmentation accuracy than lightness-order error (PLCC = 0.10, SROCC = 0.09, and KROCC = 0.06) and the average gradient of the illumination component ( PLCC = 0.35, SROCC = 0.08, and KROCC = 0.05). The proposed task-based NR-IQA metric measures the quality of illumination-corrected dermatological macroscopic photographs and correlates with segmentation accuracy.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"66 ","pages":"Article 101826"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149201045","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":"Color transfer for enhanced white blood cell classification in deep learning","authors":"Nurcahya Pradana Taufik Prakisya,&nbsp;Afiahayati,&nbsp;Dyah Aruming Tyas,&nbsp;Tri Ratnaningsih","doi":"10.1016/j.imu.2026.101828","DOIUrl":"10.1016/j.imu.2026.101828","url":null,"abstract":"<div><h3>Background and objective</h3><div>Variations in staining protocols for peripheral blood smears lead to inconsistent image coloration, posing significant challenges for automated analysis. This study evaluates the effectiveness of color transfer as a preprocessing step to harmonize blood smear image color spaces and improve white blood cell classification performance using a convolutional neural network (CNN).</div></div><div><h3>Method</h3><div>Nine color transfer algorithms were evaluated using a May-Grünwald-Giemsa-stained image as a reference. A two-stage evaluation was employed: first, assessment of the color transfer effectiveness; and second, measurement of the impact of preprocessed images on a CNN model across three backbones (ResNet-50, Inception-V3 and EfficientNet-B3).</div></div><div><h3>Results</h3><div>The findings demonstrate that improper color transfer algorithms can inherently degrade clinical feature integrity by introducing unnatural visual elements. Specifically, the color distribution transfer method causes a significant drop in classification accuracy to 72.13%. By employing Gradient-weighted Class Activation Mapping (Grad-CAM), we empirically reveal that this method generates artifacts that mislead the spatial attention of CNN away from true leukocyte structures. Conversely, adopting Monge-Kantorovich linear color mapping on ResNet-50 architecture effectively preserves semantic structures without introducing false artifacts, which facilitates more localized model attention. This method yields the highest performance, achieving an accuracy of 94.46%, specificity of 98.57%, sensitivity of 94.40% and F1-score of 94.40%, a 3.57% relative improvement over the baseline model. The results underscore the importance of selecting a robust color transfer technique for building reliable automated white blood cell classification systems that remain accurate despite variations in staining protocol.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"66 ","pages":"Article 101828"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149201046","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":"Decision support recommendations for managing Alzheimer’s dementia","authors":"Nasrullah Khan,&nbsp;Zubair Shah","doi":"10.1016/j.imu.2026.101827","DOIUrl":"10.1016/j.imu.2026.101827","url":null,"abstract":"<div><div>Knowledge graph (KG) embedding-enhanced recommender systems have shown strong potential for integrating heterogeneous healthcare data to support personalized clinical decision-making. However, Alzheimer’s Dementia (AD) care remains challenging due to the complexity of multimodal medical data and the limited ability of existing systems to incorporate user opinions and real-time contextual information. To address these challenges, this paper proposes a KG embedding-enhanced decision support system for Medication-and-dietary-pack Recommendation (KMR) in AD care. The framework integrates patient-generated feedback and disease-related contextual data into a structured healthcare KG, leverages graph neural networks to learn context-aware representations, and applies deep matrix factorization to capture latent relationships among patients, medications, and dietary components. Extensive experiments on real-world datasets demonstrate that KMR consistently outperforms state-of-the-art baselines, achieving up to a 22% improvement in NDCG and up to an 18% reduction in prediction error compared with competing methods. Moreover, KMR delivers superior recommendation quality in terms of coverage, novelty, and diversity, highlighting its effectiveness in providing accurate and clinically informed care recommendations in smart healthcare.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"66 ","pages":"Article 101827"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149155970","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":"Explainable early sepsis prediction using hybrid TCN-ransformer and retrieval-augmented clinical reasoning","authors":"Nilaksh Kashyap,&nbsp;J. Jabanjalin Hilda,&nbsp;Saman","doi":"10.1016/j.imu.2026.101817","DOIUrl":"10.1016/j.imu.2026.101817","url":null,"abstract":"<div><div>Sepsis is a life-threatening infection associated with a dysregulated host response, which is one of the major causes of death in intensive care units around the world. Early detection of the onset of sepsis is essential to enhancing patient outcomes, however the rule-based scoring systems can only be described as reactive, and do not take advantage of the rich temporal information contained within continuous monitoring data in the ICU. An explainable clinical decision support system (Early sepsis prediction) is introduced in this paper, which combines a hybrid Temporal Convolutional Network and Transformer architecture with SHAP-based explainability and retrieval-augmented clinical report generation. The proposed model works on 12-h sliding windows of multivariate ICU times series derived using the MIMIC-IV database. The proposed model which is a hybrid TCN-Transformer architecture has an AUROC of 0.777 and PR-AUC of 0.766 and it performs better than baseline models such as Random Forest, XGBoost, LSTM, and a plain TCN. SHAP gradient attribution is used to provide post-hoc explainability which provides both patient-level feature importance across both feature and temporal dimensions along with a clinical report based on verified clinical guidelines. The system is not meant to be a replacement of clinical judgment, but merely a decision-support tool to supplement it.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"66 ","pages":"Article 101817"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149155971","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":"MPANet: Cross-plane attention fusion for multi-planar brain tumour classification with auxiliary segmentation","authors":"Tanvir Ahmed,&nbsp;Rifath Mahmud,&nbsp;Syeda Anika Tasnim,&nbsp;Syma Kamal Chaity","doi":"10.1016/j.imu.2026.101825","DOIUrl":"10.1016/j.imu.2026.101825","url":null,"abstract":"<div><div>Deep learning models for brain tumour classification from MRI typically use a single viewing plane, whereas radiologists routinely examine axial, coronal, and sagittal views together. We introduce the Multi-Plane Attention Network (MPANet), a compact dual-task model that fuses features from all three planes in a single forward pass through a novel Cross-Plane Attention Fusion (CPAF) module and also produces a tumour localisation mask. Each plane is processed by its own MobileNetV3-Small encoder; CPAF lets each plane attend to the other two through query–key–value attention with squeeze-and-excitation plane weighting, and a Plane-Consistency Regularisation (PCR) loss encourages the planes to agree on their class predictions. CPAF is designed for subject-matched (Mode A) triplets, but BRISC provides no subject-level cross-plane identifiers; all results therefore use intra-class (Mode B) triplets whose planes may come from different patients, so CPAF learns class-conditioned cross-plane complementarity rather than per-patient anatomical correspondence. On the BRISC test set (540 samples, four classes), the 3.78M-parameter MPANet achieves a macro F1 of 0.9833 and a weighted F1 of 0.9796, within 1.24 percentage points of the best single-task classifier (EfficientNetB0, 0.9920), while producing tumour masks with an mIoU of 0.4853. Replacing CPAF with feature concatenation causes training to collapse (<span><math><mrow><msub><mrow><mi>F</mi></mrow><mrow><mn>1</mn></mrow></msub><mo>=</mo><mn>0.14</mn></mrow></math></span>), removing PCR reduces the combined score by 0.0278, and multi-plane fusion improves mIoU by 0.0537 over an axial-only baseline. An expected calibration error of 0.0217 indicates well-calibrated confidence on this test set but does not establish clinical applicability, which would require external validation.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"66 ","pages":"Article 101825"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149201047","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":"Development and evaluation of explainable machine learning models for colorectal cancer classification using multicenter hospital data in Thailand","authors":"Sawaporn Hinsheranan,&nbsp;Suthat Chottanapund,&nbsp;Juthaphorn Sinsomboonthong,&nbsp;Adisak Nartthanarung,&nbsp;Kritika Soha,&nbsp;Chadaporn Inta,&nbsp;Chanon Thongprayoon,&nbsp;Stephen J. Kerr,&nbsp;Sirinya Teeraananchai","doi":"10.1016/j.imu.2026.101811","DOIUrl":"10.1016/j.imu.2026.101811","url":null,"abstract":"<div><h3>Background</h3><div>Evidence on explainable machine learning (ML) for colorectal cancer (CRC) classification using multicenter routine hospital data remains limited, particularly in Southeast Asia.</div></div><div><h3>Objective</h3><div>To develop and evaluate explainable ML models for CRC classification using multicenter routine hospital data in Thailand.</div></div><div><h3>Methods</h3><div>A retrospective multicenter case–control study was conducted across 11 provincial hospitals between June 2016 and June 2017, including 504 newly diagnosed CRC cases and 997 controls. The dataset was randomly divided into training (70%) and testing (30%) sets. Model robustness was assessed using five-fold cross-validation and repeated analyses across five random seeds. Adaptive Synthetic Sampling (ADASYN) was applied to address class imbalance. Logistic Regression, K-Nearest Neighbor, Random Forest (RF), Extreme Gradient Boosting (XGBoost), and AdaBoost models were evaluated. Model interpretability was assessed using Shapley Additive Explanations (SHAP).</div></div><div><h3>Results</h3><div>Tree-based ensemble models demonstrated superior classification performance compared with traditional ML approaches. In the ADASYN with LASSO feature selection, XGBoost achieved the highest overall performance (accuracy 0.94 ± 0.006; AUC 0.98 ± 0.002; sensitivity 0.91 ± 0.005; specificity 0.95 ± 0.007), while RF showed comparable discrimination performance (accuracy 0.93 ± 0.003; AUC 0.98 ± 0.001). SHAP analysis identified hypertension, blood pressure measures, fasting plasma glucose and behavioral indicators as influential features associated with case–control classification.</div></div><div><h3>Conclusions</h3><div>Tree-based ensemble models demonstrated strong and robust performance for CRC classification using routine hospital data. SHAP analysis identified key socioeconomic, metabolic, clinical, and behavioral features associated with classification. External and prospective validation studies are needed to assess generalizability and support future implementation.</div></div>","PeriodicalId":13953,"journal":{"name":"Informatics in Medicine Unlocked","volume":"65 ","pages":"Article 101811"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148952495","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}