Adaptive Graph Cuttweedie Regressive Deep Convolutional Learning For Kidney Cancer Classification With CT Images

Abstract

Medical image processing is the medical image analysis for improving quality through extracting the features for efficient classification. Image classification is an important one for premature recognition as well as treatment of kidney disease. Many researchers carried out their research on automated detection and classification of kidney cancer. However, the conventional classification methods has limited multi-modal incorporation, lack of generalization and reduced analyzability. In addition, the morphological characteristics are not identified and resulted in higher risk of misdiagnosis. In order to address these issues, Adaptive Graph Cut Tweedie Regressive deep Convolutional Learning (AGCTRdCL) Model is introduced. Major aim of AGCTRdCL is to perform automated recognition and classification of kidney cancer. AGCTRdCL Model performed four process for kidney cancer diagnosis. Initially, number of CT kidney images is considered as an input at input layer. After that, Adaptive Gaussian Image Preprocessing task is carried out in hidden layer 1 for noise filtering, contrast enhancement and normalization to enhance quality of kidney CT images. Then, Graph Cut Segmentation is carried out in hidden layer 2 for segmenting the pre-processed images into number of segments. Segmented regions of CT images are sent to the hidden layer 3. In that layer, Tweedie Regressive Feature Extraction (TRFE) process and classification is carried out in AGCTRdCL Model. Tweedie Regressive Feature Extraction is used to extract the shape, color and texture of segmented regions. With the extracted features, softplus activation function is used for performing efficient classification of kidney cancer. Finally, the output layer displays the classification results. The fine tuning process of the deep convolutional learning is carried out through gradient descent concepts. This in turn, efficient kidney cancer classification is carried out with higher accuracy and minimum time complexity. The performance parameters are kidney disease classification accuracy, kidney disease classification time, peak signal-to-noise ratio, FE accuracy, Pre, Rec , f1-score.