Histopathological Cancer Detection Using Pre-Trained Models

Early and accurate cancer identification is essential for ideal treatment planning. For cancer identification histopathological mages, the images obtained from biopsy tissue are considered to provide a more accurate prediction than other methods. Manual histopathologic analysis provides significant issues, such as image complexity and variability. This study delves into the utilization of deep learning models to analyze histopathological images for the purpose of cancer diagnosis. Also, the concepts of transfer learning and ensemble learning are leveraged to enhance the predictive accuracy of cancer detection. The performances of CNN architectures are evaluated based on the assessment parameters Accuracy, Sensitivity, Specificity, and F1-Score. The ensemble learning model comprising ResNet-50, VGG- 16, VGG-19, and DenseNet-121 performs better than other models. In addition, the eXplainable Artificial Intelligence technique, LIME, provides insights into the factors deriving the predictions. The saliency map and feature indices present a comprehensive overview of how the model utilizes features to generate the predicted output.