Lung and Colon (L&C) cancers are fatal diseases that can develop in several organs simultaneously and, in certain situations, endanger human life. While the concurrent development of these two forms of cancer is very unlikely, delayed diagnosis greatly increases the likelihood of metastasis between the affected organs. To effectively treat certain types of cancer, histological diagnosis is essential. Traditionally, doctors had to go through a lengthy and laborious process to review histological images and diagnose cancer cases; however, with the new technology options, this process may now be completed much faster. In this study, a hybrid Deep Learning (DL) model combining an attention mechanism and a multipath network was employed to classify histological images of L&C tumors. To focus on the most important characteristics and disregard the less important ones, an attention mechanism was employed. Data is sent over numerous channels in a multipath network, which then converts each channel and combines the output from all of the branches. Simplified, the multipath network is just like grouped convolution. The LC25000 dataset was utilized, which included five categories of histopathological images. Among these categories were two for colon cancer and three for lung cancer. Several popular DL models, including ResNet-50, VGG-16, and AlexNet, were used to compare the effectiveness of the proposed model. The proposed approach showed the best performance in terms of accuracy (99.2%), specificity (99.12%), sensitivity (99.28%), precision (99.12%), and F1 score (99.2%), according to the experimental findings.

Medical Image, Deep Learning, Accuracy, Histopathology, Cancer, Pre-process