Automated Classification of Kidney Tumours Using Deep Convolutional Neural Networks

BACKGROUND: Kidney tumour (KT) represents a major global health burden, being the seventh most frequently diagnosed cancer worldwide. Early detection is crucial for the success of treatment, and it typically relies on imaging techniques such as computed tomography (CT), which is time-consuming and tedious. METHODS: Towards detecting and classifying KT, we leverage two deep learning models (EfficientNetV2-B3 and ResNet50) as dual backbones for feature extraction. Additionally, our method further utilises dual efficient channel attention module (DECA) for effective cross-model feature fusion. RESULTS: Our method achieved excellent results, compared to alternative feature fusion strategies like cross-attention or weighted-sum. It reaches a 98.80% accuracy and a 98.28% F1 score on the CT Kidney dataset. CONCLUSIONS: Our proposed model is a powerful network consisting of the dual backbone integration and DECA, achieving superior performance in the task of classifying renal tumour CT images.