Incorporating Residual Connections into a Multi-channel CNN for Lung Cancer Detection in Digital Pathology

Early detection is vital for the early treatment of lung cancer (LC), resulting in a reduced mortality rate. This research focuses on not only detecting LC with the help of computer vision techniques, but also reducing patient cost and time. In this research, a robust model was proposed on the 2019 IQ-OTH/NCCD LC dataset by incorporating multichannel advantages into a convolutional neural network integrated with the residual connection. A dual channel with a kernel size of 5*5 and a swish activation function has been employed in multichannel features. InceptionResNetV2 was selected as the base model in this experiment. Multichannel techniques extract richer features from diverse aspects of the input data while residual connections diminish the gradient problem, resulting in an optimized and efficient training process. Our proposed model outperformed all existing models including the SOTA model, with an accuracy of 89.95%, precision of 91.42%, recall of 88.84%, F1 of 89.68%, and specificity of 95.98%.