Use of Client-Side Machine Learning Models for Privacy-Preserving Healthcare Predictions - A Deployment Case Study

Machine learning (ML) and deep learning (DL) models in healthcare traditionally rely on server-centric architectures, where sensitive patient data is transmitted to external servers for processing via frameworks like Flask, raising significant privacy concerns. This work demonstrates a privacy-preserving approach by executing healthcare prediction models entirely within the web browser.

Our approach leverages existing browser-based machine learning and deep learning technologies such as TensorFlow.js and ONNX Runtime Web, along with direct JavaScript implementations, to ensure all computations remain on the client side. We showcase three implementation strategies based on model complexity: direct JavaScript implementation for simple equation-based models, ONNX-based conversion and execution for medium-complexity models like Random Forest and finally TensorFlow.js deployment for complex deep learning models such as Optimized Convolutional Neural Networks.

Our results indicate that client-side deployment is both feasible and effective for healthcare prediction models, preserving original performance metrics while offering substantial privacy benefits.

This approach guarantees patient data never leaves the user's device, eliminating risks associated with data transmission and making it particularly advantageous in healthcare settings where data confidentiality is critical, while also supporting offline functionality.