Classifying Kidney Tumor via Adaptive SE‐ResNeXt With Mutli‐Novel Loss Function and Renal Mass Segmentation Using Pyramidal Attention‐Based R2Unet++ Model

Kidney tumors are one of the prevalent types of tumor globally and it has become a significant health concern. It is the most frequent type of urological tumor. While kidney tumors can present in various forms, the majority of the tumors are malignant. Hence, early detection is needed to reduce the mortality rates, develop preventive measures to minimize their impact and provide effective treatments. Compared to traditional, labor-intensive diagnostic process, deep learning-based automatic detection algorithms can reduce diagnosis time, lower costs and improve accuracy. In this research, a kidney cancer detection model using deep learning is introduced to speed up the diagnosis process with high efficiency. At first, the input images are taken from benchmark sites. Then, segmentation is performed on the input images using developed Pyramidal Attention-based Recurrent Residual Unet++ (PA-R2Unet++) model. The development of a tumor is identified in the early stages by this accurate segmentation procedure. Then, the segmented images are used in the classification procedure and it is accomplished via Adaptive Squeeze-And-Excitation-ResNeXt with Novel Multi-Loss Function (ASE-RX-NMLF). In addition, an Enhanced Arbitrary Value-based Flamingo Search Algorithm (EAV-FSA) is used to tune the parameters from the suggested ASE-RX-NMLF for improving the classification performance. Classifying kidney tumors based on their characteristics helps to enhance the effectiveness of therapies. The efficacy of the developed method is verified by comparing the final results of existing models.