Colon Cancer Detection Using YOLOv5 Architecture

Cancer is a disease where the cell divides exponentially and damages the body tissues. Early diagnosis of cancer can help the person from witnessing complicated treatments and indeed death. According to a study, early detection of curable cancer can save a person's life up to 99%. The exponential cell division in other words is known as tumor development in the body. The condition where the tumor has not started to spread to other body parts is known as benign. Utmost of the cysts in the colon are benign. The condition when the tumor became cancerous it is known as adenocarcinomas. Colorectal cancer is the 2nd leading cause of cancer death. It is found that human evolved cell slides aren't accurate enough to detect cancer. It takes about 1 or 2 weeks to get the result from a laboratory. However accurate and quick discovery of benign/adenocarcinomas is essential to detect colon cancer. A convolutional neural network increases the standard of the diagnosis by giving an accurate and quick result. The deep convolutional neural network can best serve the purpose to detect colon cancer. In the proposed system, yolov5 architecture is used to descry cancerous cells. A custom dataset of about 2000 images is used to train the model. The proposed model helps to separate the tumor cell into benign or adenocarcinomas. The training of the model in yolov5 is processed using PyTorch. The time taken to train the Colon Cancer Detection model is 78 minutes. The model is tested with 140 images and the accuracy is 99.28%. The latency for each image is 0.031 seconds. This proves that the proposed method has better efficiency when compared to the traditional laboratory system of detection of colon cancer. The proposed work may assist the clinicians in reducing the missed diagnosis during endoscopy.