Fractional Stochastic Gradient Descent Trained SM-Segnet-Based Segmentation And Capsule Neural Network For Colon Cancer Detection

Abstract

Cancer that forms in the epithelial cells of the large intestine which then proliferate to the neighboring regions is called colon cancer. Early diagnosis is vital for efficient treatment and for the recovery from this disease. Detection of colon cancer at the preliminary stages can significantly assist physicians in decision-making and thus decrease the hardships. Accurate detection of colon cancer could be attained by automatic models that process the medical image to detect cancerous growth. The prevailing techniques developed for colon cancer detection that depend on Deep Learning (DL) models necessities more computational ability as well as resources. This paper proposes a novel segmentation technique that uses the Fractional Stochastic Gradient Descent (FSGD) to train the Squeeze M-SegNet (SM-SegNet). The SM-SegNet is utilized for segmenting the required region of the colonoscopy image that is preprocessed by the Non-Local Means (NLM) filter for noise removal. Finally, the Capsule Neural Network (CapsNet) is used for detecting colon cancer. The developed methodology is analyzed by considering metrics like specificity, sensitivity, and accuracy. The experimental result illustrates that the developed method recorded an accuracy of 0.918 which is comparably superior to the prevailing systems, and sensitivity and specificity recorded are 0.907 and 0.927 respectively.