Optimized Feature Selection and DenseNet-121 Based Deep Learning Model for Predicting Ulcerative Colitis Cancer

Ulcerative Colitis (UC) is marked by persistent inflammation of large intestine in individuals with a genetic predisposition who are exposed to certain environmental risk factors. The cause of the disease is uncertain to predict and ranges from mild to severe. Early prediction and accurate diagnosis is essential to improve patient outcome. This paper proposes an optimized feature selection and Deep Learning (DL) based DenseNet121 classifier for predicting UC cancer. Non-Local Means (NLM) preprocessing based filter is applied in the collected data to remove noise and to enhance the resolution of the image. The preprocessed data is followed by mean shift algorithm that segments the regions of the medical image. Local Binary Pattern (LBP) is utilized to extract essential features from image. DenseNet 121 classifier is employed to predict cancerous and non-cancerous lesions. The key parameters of the classifier is finely tuned by Particle Swarm Optimization (PSO) algorithm to enhance classifier accuracy. The efficiency of the proposed model is implemented in the Python software and evaluated using performance metrics such as accuracy of 94 %, precision of 95 %, recall, and F1-score of 94 % respectively ensuring robust and automated diagnostic model for UC cancer prediction.