Modeling and Detection of Colorectal Cancer ImagesUsing Transfer Learning and Convolutional NeuralNetworks (VGG۱۶)

Colorectal cancer is recognized as oneof the most common cancers of the present era.Early detection of this type of cancer cansignificantly facilitate doctors' decision-makingand reduce their workload. Recently, the successof artificial neural networks in identifying andclassifying disease lesions has encouragedresearchers to use this method for processingmedical images. The present study wasconducted with the primary aim of presentingan artificial neural network algorithm to detectcolorectal cancer in medical images. Morespecifically, colorectal cancer is a commonmalignancy, and accurate tissue analysis is vitalfor diagnosis and treatment planning. In thisstudy, we use transfer learning and the VGG۱۶convolutional neural network to classify tissuesin histopathological images of colorectal cancer.Using the Kather_texture_۲۰۱۶ dataset, whichcontains ۵,۰۰۰ histology images classified intoeight types of tissues, we preprocess andaugment the data to increase the model'sgeneralization. Our approach integrates a pre-trained VGG۱۶ model that is fine-tuned with additional custom layers to extract robustfeatures and achieve high classificationaccuracy. The model is trained and validatedusing a precise split of training, validation, andtest sets. Our results show significantperformance with ۹۲% accuracy and a Cohen'skappa score of ۰.۹۱, indicating strong agreementwith the actual labels. This study emphasizes thepotential of deep learning and transfer learningin advancing the accuracy of colorectalhistopathological analysis, contributing to morereliable and efficient diagnostic processes.

کلیدواژه ها:

Colorectal Cancer ، Artificial NeuralNetwork ، Colorectal ، Rectum ، Medical Imaging