Bsingstad/GMC2025

George Moody Challenge 2025 - classification of Changa disease # Cha-Cha-Chagas: Auxiliary Pretraining and Fine-Tuning Across Heterogeneous Datasets for ECG-Based Chagas Disease Detection This repository contains the **Python implementation** for our entry in the **George B. Moody PhysioNet Challenge 2025**, titled *"Auxiliary Pretraining and Fine-Tuning Across Heterogeneous Datasets for ECG-Based Chagas Disease Detection"*. Our submission was developed by the **Cha-Cha-Chagas** team and builds upon the official PhysioNet Challenge 2025 Python example repository, expanding it with deep learning–based architectures, auxiliary pretraining, and fine-tuning strategies for improved ECG-based Chagas disease detection. --- ## 🧠 Overview Chagas disease (American trypanosomiasis) is a parasitic infection caused by *Trypanosoma cruzi*, soemtimes leading to **chronic Chagas cardiomyopathy (CCC)**. Detection from ECG signals remains challenging due to the scarcity of high-quality labeled data. Our approach investigates whether **auxiliary pretraining on weakly labeled ECG data** (from the large CODE-15% dataset) can improve downstream Chagas detection when **fine-tuned on datasets with stronger labels** such as SaMi-Trop (serologically confirmed positives) and PTB-XL (assumed negatives). Despite the hypothesis, experiments revealed that this pretraining strategy **did not outperform conventional supervised training**, underscoring the importance of dataset balance, label reliability, and domain similarity in multi-dataset ECG modeling. --- ## 📁 Repository Structure * `train_model.py` — Wrapper for training. * `run_model.py` — Wrapper for inference. * `evaluate_model.py` — Used for local validation with the official PhysioNet evaluation code. * `team_code.py` — **Main implementation** containing: * Deep neural network architecture (`Net1D`). * Dataset handling (`ECGDataset`). * Auxiliary pretraining and fine-tuning logic. * Model saving/loading utilities. As instructed by the Organizers, we did **not** modify the official Physio …