Seizing the opportunity presented by the George B. Moody PhysioNet Challenge 2025, our team medex has devised a novel ResNet18-based framework to identify Chagas disease from 12-lead ECGs. To address the challenges posed by heterogeneous data sources (CODE-15%, SaMi-Trop, PTB-XL), which feature variable sampling rates (100-500 Hz) and durations (7.3-10.2 s), we standardize the ECG signals by resampling them to 500 Hz. This process leverages the Fourier transform to reconstruct signals in the frequency domain, ensuring that key spectral characteristics are preserved while achieving a uniform sampling rate. Following resampling, we extract fixed-length 2500-sample segments (equivalent to 5-second windows) through randomized cropping, enhancing the model's ability to generalize across varying data distributions. Our model builds on the ResNet18 architecture, modified specifically for 1D ECG signals. These adjustments include redesigned convolutional layers and a cross-lead feature fusion mechanism to effectively capture inter-lead dependencies. On the hidden validation set, our approach achieved a challenge score of 0.38, showcasing its potential to advance the automated detection of Chagas disease.