NeuroGuard

Overview

NeuroGuard (StrokeRiskCalc) is a high-recall clinical stroke risk estimation platform that analyses 14 physiological and lifestyle parameters to identify individuals at risk of stroke. It achieves 95.62% clinical sensitivity through dynamic decision threshold tuning, and uses SHAP explainability to deliver personalised, actionable clinical recommendations — not just a score.

The Problem

Traditional stroke risk classifiers use a standard 50% probability boundary, which yields an unacceptable ~5.8% recall for detecting actual stroke cases. Most people are unaware of their risk until a critical event occurs, and existing consumer tools lack clinical rigour, privacy compliance, or explainability.

The Solution

A XGBoost classifier with monotonic constraints ensures that increasing physiological risk factors (Age, BP, BMI, Glucose, Cholesterol) never mathematically decrease the predicted risk. A serverless Firebase backend processes patient data with strict PDPA/APP privacy compliance. An OCR pipeline automatically extracts clinical biomarkers from scanned medical reports. SHAP values translate raw ML decisions into personalised clinical rationale — explaining exactly why the risk score was assigned.

Skills & Technologies

Tools Used

Key Achievement

95.62%

Clinical Sensitivity

All Projects