Logo image
Back
Trustworthy AI for Early Dementia Detection: Robust Feature Masking and Clinical Interpretability
Conference proceeding   Open access

Trustworthy AI for Early Dementia Detection: Robust Feature Masking and Clinical Interpretability

Konstantinos Georgiou, Longjian Liu, Hairong Qi and Xiaopeng Zhao
IEEE/ACM Conference on Connected Health: Applications, Systems and Engineering Technologies (Online), pp 279-283
24 Jun 2025
DOI: https://doi.org/10.1145/3721201.3721410
Featured in Collection :   Research Supported by Drexel Libraries' OA Programs
url
https://doi.org/10.1145/3721201.3721410View
Published, Version of Record (VoR)Open Access via Drexel Libraries Read and Publish Program 2025CC BY V4.0,  Open

Abstract

Data integrity Data models Decision making Early Dementia Detection Feature extraction Feature Importance Interpretability Medical services Neural ODE Resilience Robustness SHAP Training Transformer Transformers Dementia
Early dementia detection is pivotal for timely clinical interventions that can delay cognitive decline and improve patient quality of life. However, many machine learning models exhibit fragility by depending too heavily on a small subset of features, reducing both robustness and interpretability-particularly when data quality or completeness varies. To address this issue, we introduce a random feature masking strategy that deliberately obscures some inputs during training to encourage broader and more balanced feature usage. This approach improves resilience to missing or degraded data and enhances clinical trustworthiness by producing more interpretable outputs. We demonstrate our method on the Women's Health Initiative Memory Study (WHIMS) dataset using Transformer and Neural Ordinary Differential Equation (Neural ODE) models under two dataset configurations. Evaluations based on Macro F1 and SHAP-based feature importance show that random masking notably improves both robustness and interpretability, reducing dependence on dominant features while maintaining or boosting predictive performance. These results highlight the clinical potential of random feature masking in creating reliable, interpretable models for early dementia detection, pointing toward more trustworthy AI-driven healthcare solutions.CCS Concepts* Computing methodologies → Supervised learning by classification.

Metrics

1 Record Views

Details

InCites Highlights

Data related to this publication, from InCites Benchmarking & Analytics tool:

Collaboration types
Domestic collaboration
Web of Science research areas
Computer Science, Interdisciplinary Applications
Engineering, Biomedical
Medical Informatics
Logo image