AI-Driven Hybrid CNN-Transformer Model for Detecting Cardiotoxicity in Cancer Chemotherapy Patients

Abstract

irreversible cardiac dysfunction, making early and accurate detection crucial for 
effective patient management. This work presents an AI-driven hybrid CNN
Transformer framework for automated CIC detection from echocardiography 
data, formulated as a binary classification task between cardiotoxicity and non
cardiotoxicity cases. The model integrates convolutional neural networks for 
discriminative spatial feature extraction with Transformer layers to capture 
global and temporal cardiac motion patterns, enabling robust representation 
learning without reliance on handcrafted features. Experimental evaluation 
demonstrates an overall detection accuracy of 92.2%, achieving competitive 
performance against recent deep learning approaches, including CoreEcho and 
R3D Transformer models. The proposed framework also exhibits improved 
generalization and computational efficiency, supporting scalability across 
diverse clinical settings. Furthermore, the model’s automated design reduces 
observer dependency and enhances diagnostic consistency. These results 
highlight the potential of hybrid CNN–Transformer architectures for non
invasive, reliable, and early cardiotoxicity screening in real-world oncology 
practice.