Chagas disease affects 6-7 million people globally with severe underdiagnosis rates exceeding 95% in endemic re- gions [1,4]. Due to the complexity and challenges of sero- logical testing, a deep learning model that uses a stan- dard ECG for the diagnosis of the disease can be really helpful for the diagnosis of the same. This paper presents the work of our team (MCV UVIC) for the Physionet chal- lenge 2025 that uses a robust deep residual neural net- work with signal processing for 12-lead ECG analysis. The architecture implements hierarchical temporal feature extraction through residual blocks with progressively re- fined temporal resolution, combined with spatial dropout (p = 0.1) and L2 regularization (λ = 10−4). Our pre- processing techniques utilize percentile-based normaliza- tion (P01-P99 clipping) and adaptive resampling to 400 Hz, employing sinc interpolation. For extreme class im- balance (P (Y = 1) < 0.05), we have implemented syn- thetic minority oversampling with controlled noise injec- tion and temporal augmentation. The system achieves the physionet evaluation score of 0.272 while maintain- ing computational efficiency (< 100 ms inference). Key innovations include: (1) Clinically-aware preprocessing pipeline (2) Progressive temporal feature extractions, and (3) Challenge-optimized class imbalance mitigation .