Analysis of CT images for detection of Colorectal Cancers using Hybrid Artificial Neural Networks and Fire Fly Algorithm

Abstract The longest part of the large intestine or colon is called as rectum. The uncontrolled division of cells in the colon forming a polyp at initial stage is called as colorectal cancer. Initially the polyp formed may be benign and later it may turn out to be malignant. The advancements in the field of medicine has made the screening techniques to be an effective tool in identification of the colon cancer. Early detection of the colon cancer may lead to complete cure of the disease normally, but it is very difficult to diagnose at the start. It is because, it does not show any symptoms at initial stages. Colonoscopy, a type of computed tomography (CT) is usually recommended for the patients to detect the colon cancer and it is usually a painful test conduct to identify the disease. To relieve the patients from the suffering, image processing algorithms like Weighted Adaptive Scalable Invariant Transform (WASIT) is used for extraction of features like location, orientation and scale are used as inputs to train the Artificial Neural Network (ANN) using Back Propagation Algorithm (BPA). The optimal set of weights are obtained by adjusting the weights for BPA hybrid with Genetic Algorithm (GA) and Fire Fly Algorithm (FFA). The Likelihood Ratio (LR+) is found to optimal for the BPA tuned with FFA and is inferred to be in the range of 1.5 to 4 which varies with a deviation of ±0.4% from the nominal value.