Clinically guided subgroup-aware framework for dental implant risk stratification using expert-informed feature selection and cross-group knowledge transfer
Year: 2026
Type: article
Abstract: Dental implant outcome prediction is challenging due to the interplay of systemic health variables, oral conditions, and anatomical parameters. Existing machine learning approaches often treat patient populations as homogeneous and optimize a single global feature subset. This approach limits sensitivity to subgroup-specific risk patterns, reduces interpretability, and weakens the detection of high-risk failure cases. This study proposes TBLPSO, a team-based learning particle swarm optimization framework that integrates clinically guided subgrouping, expert-informed initialization, reliability-aware weighting of expert input, and cross-group feature collaboration. Patients are stratified into four clinically meaningful subgroups: Low Risk, Local Risk, Systemic Risk, and Complex Case, using a real-world European dental implant dataset comprising a total of 55 clinical features. The dataset was chronologically divided into training and testing sets, and five-fold temporal cross-validation was applied to the training set. Each subgroup is optimized independently while selectively sharing predictive features across groups. This enables the model to capture both subgroup-specific and generalizable risk factors. Extensive comparisons with PSO variants, wrapper-based methods, evolutionary algorithms, and standard classifiers demonstrate that TBLPSO consistently outperforms these methods. The framework attains an accuracy of 96.1% while maintaining compact subgroup-specific feature subsets (5, 6, 9, and 12 features, respectively). The proposed cross-group communication mechanism enables structured knowledge transfer, allowing generalizable predictors to propagate across subgroups while preserving context-specific features, supporting future investigation toward clinical decision-support applications.
Source: Scientific Reports
Institution: Xi'an Jiaotong University
Language: English
Subfield: Health Information Management
Field: Health Professions
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $2,690
OpenAlex ID: W7223104333