Chemotherapy-induced cardiotoxicity (CIC) remains a significant clinical concern, often leading to long-term cardiac complications in cancer patients undergoing treatment. Accurate and early prediction of cardiotoxic events can enable timely intervention and improved patient outcomes. This project presents a multimodal deep learning framework that integrates convolutional and sequence-based neural architectures for the binary classification of cardiotoxicity outcomes, namely Cardiotoxicity (CTRCD) and Non-Cardiotoxicity (NO_CTRCD). The proposed system leverages Temporal Dynamic Imaging (TDI) data as the primary modality. A Convolutional Neural Network (CNN) is first employed to extract robust spatial feature representations from the TDI images. These extracted features are then passed to a sequential learning module, evaluated through three configurations: CNN + LSTM, CNN + GRU, and CNN + Transformer. Among these, the CNN + Transformer architecture demonstrated superior performance, effectively capturing both local image-based features and long-range temporal dependencies. The model architecture involves concatenating CNN-derived feature vectors, processing them through a Transformer encoder, and feeding the combined representation into a dense layer for final classification. Comprehensive experiments were conducted to evaluate the predictive capability of each model configuration. The CNN + Transformer model achieved the highest accuracy and F1-score, highlighting its potential as a reliable diagnostic tool for early detection of CIC. This multimodal framework shows promise in enhancing precision cardiology and supporting oncologists in making risk-informed decisions during chemotherapy planning.

Keywords: Chemotherapy-Induced Cardiotoxicity, Multimodal Deep Learning, CNN, Transformer, TDI Imaging, Sequential Modeling, LSTM, GRU, Feature Fusion, Cardiotoxicity Prediction, CNN-Transformer Architecture, Biomedical Signal Analysis, Clinical Risk Stratification, Deep Learning in Cardiology.