Improved Early Kidney Oncocytoma Identification in CT Scans using Hybrid CNN-RNN/Attention Models

Aim: The aim of this paper is to improve the accuracy of initial kidney oncocytoma diagnosis on CT scans using a Hybrid deep learning model based on Convolutional Neural Networks (CNN) and processes. The performance of the proposed model is compared to Recurrent Neural Networks (RNN) with attention and standard CNN-based techniques. Materials and Procedures: Two groups are considered in this research: Group 1 uses a standard CNN model with 140 samples, and Group 2 is a Hybrid CNN-RNN/Attention model with 140 CT scan samples for training and validation. A 95% CI is kept, the threshold is at 0.05%, and the power value is at 80%. Result: Consequently, the hybrid CNN-RNN/Attention model's accuracy is significantly higher than that of the standard CNN model. While the range for the CNN model was 75.3% to 85.1%, the range for the hybrid model was 88.2% to 96.5%. At a sensitivity of 0.89 and a specificity of 0.92, when the significance level was 0.003, the hybrid model was at its best. Conclusion: In contrast to traditional CNN models, the hybrid CNN-RNN/Attention model demonstrated superior accuracy in kidney oncocytoma early detection in CT images. The results suggest that diagnostic accuracy of oncocytoma detection can significantly be enhanced using hybrid deep learning models.