Colon cancer classification and detection by novel CMNV2 model and methods of deep learning

Colon cancer is the leading cause of death among cancers. Colon cancer, commonly referred as colorectal cancer (CRC), has two common names. Adenomas are a sequence of benign stages that colon cancer undergoes after starting in a healthy intestinal cell. Reducing the possibility of treatment failures and consequences requiring early identification of cancer, our study started with pre-processing digital histopathological images using transfer learning and deep learning (DL) models. A total of 12 models were evaluated and compared in this research, including standard methods like MobileNet, NASNetMobile, DenseNet121, DenseNet169, InceptionV3, DenseNet201, ResNet152V2, EfficientNetB0, InceptionResNetV2, Xception, EfficientNetV2M and our proposed model. The proposed hybrid model was developed by combining the convolutional architecture for fast feature extraction (CAFFE) framework with the modified MobileNetV2 (MMNV2) framework named as 'CAFFE+MMNV2 (CMNV2).' To increase performance, classification, detection and prediction accuracy in colon cancer, this study presents a novel CMNV2 model created by an additional 5-layer pre-trained model for feature extraction from images using DL, deep neural network (DNN). The suggested CMNV2 model outperformed the remaining 11 existing methods for colon cancer classification and detection with a 0.001 learning rate. Feature extraction from histopathological images to classify and detect colon adenocarcinoma (COAD) and colon benign tissue (COBT) is implemented by a Python model using a dataset consisting of 10,000 images. The suggested model significantly outperformed other existing methods, obtaining 99.95% accuracy and other high-performance criteria including 100% recall, 99.90% precision, 99.95% f1-score and 0.05% error rate, when using fewer parameters.