As part of the "Detection of Chagas Disease from the ECG: The George B. Moody PhysioNet Challenge 2025", we employed two lightweight convolutional neural network architectures based on modified VGG and ResNet models to detect Chagas disease from 12-lead electrocardiogram (ECG) signals. Our preprocessing pipeline included resampling to 128 Hz, fixed-length cropping or zero-padding, bandpass filtering (1–40 Hz), and channel-wise normalization. To address pronounced class imbalance, we employed a balanced sampling strategy during training and used Monte Carlo Dropout at inference for uncertainty estimation. Evaluated via 5-fold cross-validation, the best model (LiteVGG-11) achieved an AUROC of 0.842±0.009 and a recall of 0.725±0.018, but low precision (0.066±0.002), resulting in a challenge score of 0.410±0.009. These results highlight both the potential and challenges of automated Chagas disease detection from ECGs in highly imbalanced datasets. On the hidden validation set provided by the challenge, our team, AIMED, obtained a final challenge score of 0.372, ranking 31st overall and 10th among unique team entries, demonstrating the promise of lightweight models for scalable and portable screening solutions in resource-limited settings.