TrustECG

Tech Stack

The Problem

• Specialist cardiologists were the only qualified reviewers of ECG data, creating severe screening bottlenecks.
• Rural and resource-limited healthcare settings lacked access to cardiologists for routine cardiac screening.
• Manual ECG reading averaged 10–20 minutes per patient, creating unsustainable workloads at scale.
• Subtle arrhythmia patterns were frequently missed under high consultation volumes and clinician fatigue.
• High cardiac screening costs prevented population-level preventive cardiology programmes from scaling.

Our Solution

• Developed a deep learning model combining PyTorch and TensorFlow for automated ECG classification.
• Trained on large-scale annotated ECG datasets covering normal sinus rhythms and multiple arrhythmia variants.
• Built an ECG signal preprocessing pipeline for noise denoising, baseline correction, and feature normalisation.
• Implemented confidence scoring to automatically flag borderline cases for specialist cardiologist review.
• Designed a clinical-grade inference pipeline producing structured classification reports per ECG reading.

Impact

Automated ECG screening, enabling faster cardiac risk assessment for large patient populations.