Machine Learning-Driven Clinical Decision Support Systems For Healthcare Optimization

Abstract

In the era of digital healthcare, the exponential growth of sensitive medical data presents a dual challenge: leveraging data for AI-driven insights while preserving patient privacy. Traditional centralized AI models pose significant risks related to data breaches, regulatory compliance, and trust among data providers. To overcome these issues, this paper suggests a federated learning (FL) model that incorporates blockchain technology to facilitate the implementation of secure and decentralized training of models in concert among various healthcare facilities. This study aims at creating a system whereby local models are trained on-site using clinical, research and behavioral data, and at most only model parameter is shared and not the raw data. A consensus mechanism to guarantee integrity and traceability of model updates is a blockchain-based method, in which validated updates are permanently recorded with the help of smart contracts. The methodology has been organized into three levels, namely, data extraction and storage, data management, and data application. The findings show that the proposed FL-CNN-HMChain model outperforms traditional CNN and VGG16 architectures, achieving AUC scores above 85% across all tasks, with a maximum of 98.9% on Organ AMNIST. This approach demonstrates both high accuracy and strong privacy preservation, paving the way for scalable and ethical AI deployment in healthcare.