Chagas disease, caused by Trypanosoma cruzi, is a parasitic illness that can lead to chronic cardiomyopathy, and may be fatal in severe cases. Compared to conventional serological tests, electrocardiogram (ECG) screening is more cost-effective and rapid. In this paper, a supervised deep learning method for ECG-based detection is proposed. The training set included all samples from Sami-Trop and PTB-XL datasets, along with randomly downsampled negative samples from CODE-15% dataset. Based on characteristic ECG manifestations of Chagas disease, four leads (II, V1, V4, V6) were selected. The signals were preprocessed through resampling to 250 Hz and bandpass filtering (1-45 Hz), then fed into a hybrid 1D-VGG16-BiLSTM model to extract both short-term and medium-term temporal features. To address severe class imbalance, focal loss was employed as the cost function with AU-PRC as the primary metric. A bagging ensemble of five models with soft voting was implemented for final prediction. Experimental results showed a mean challenge score of 0.464 (±0.030) in 6-fold cross-validation, with an official score of 0.439 achieved by GDMUBME_ECGGroup. The proposed method shows promising potential for Chagas disease detection.