Detection of Cancer Cells Using Convolutional Neural Networks (CNNs)

The fastest spreading dieses among the cancer dieses throughout the world is the Breast cancer. Identifying it at the prior stage will be helpful for effective treatment. Convolutional Neural Networks (CNNs) is an efficient technique which has demonstrated significant ability in classifying and recognizing patterns in medical images. In this research paper we have explored the application of deep learning models for detecting cancer cells from histopathological images, and focused on various architectures, training methods, and assessment measures. Here we specifically evaluated the functioning of four well-known models of CNN comprising AlexNet, VGG16, InceptionV3 and, ResNet50 by employing the dataset from BreakHis for binary classification of the histopathological images into Malignant and Benign classes. Each model is fine-tuned and trained through transfer learning. Here we used cross-entropy loss with Adam optimizer for binary classification, through which the learning rate is adaptively adjusted, and facilitating efficient handling of large datasets. In this research work for training our models we applied Stratified k-fold cross-validation approach for ensuring stability and generalizability. The effectiveness and performance of the models are assessed with the help of Accuracy, Precision, & Recall, and F1-Score. Comparative analysis indicates that ResNet50 outperforms the other models in detecting cancer cells. Finally, we discussed the potencies and weakness of the applied models for offering valuable perceptions into their efficacy for medical image classification tasks.