ChemoCardioNet: An Explainable Multimodal Transformer for Early Prediction of Chemotherapy-Induced Cardiotoxicity.

Area of Science:

• Cardio-oncology
• Artificial Intelligence in Medicine
• Biomedical Engineering

Background:

• Chemotherapy-induced cardiotoxicity (CIC) is a significant concern for cancer survivors.
• Current monitoring methods often miss early, subclinical cardiac damage.
• Early detection of CIC is crucial for effective patient management and improved outcomes.

Purpose of the Study:

• To develop and validate ChemoCardioNet, an explainable multimodal deep learning framework.
• To predict cardiotoxicity risk before clinical manifestation using integrated patient data.
• To enhance early risk identification and cardio-oncology monitoring.

Main Methods:

• Developed ChemoCardioNet, a multimodal deep learning framework integrating ECG, echocardiography, clinical variables, and biomarkers.
• Employed modality-specific encoders (1D CNN transformers, CNN-ViT, MLPs) and a cross-attention fusion transformer.
• Utilized SHAP analysis for feature-level explainability of cardiotoxicity risk scores.

Main Results:

• ChemoCardioNet achieved an AUC of 0.92 in a cohort of 1,200 patients.
• The framework outperformed single-modality and traditional machine learning approaches by 8-12%.
• Cardiotoxicity was predicted approximately two cycles earlier than detectable echocardiographic dysfunction.

Conclusions:

• ChemoCardioNet demonstrates high accuracy in predicting chemotherapy-induced cardiotoxicity.
• The model offers superior performance and earlier detection compared to existing methods.
• This framework supports proactive cardio-oncology care and personalized cancer treatment strategies.