LungSE-Net: Enhanced Lung Cancer Diagnosis via Lightweight CNN Model using Histopathological Images

In pathological diagnosis process classification of histopathological images is an essential step. To improve the diagnostic facilities, it becomes very important to use deep learning as a method to diagnose disease as it can capture complex features from the image. Our study aims to develop a convolutional neural network model for lung cancer classification, which is designed to be lightweight in terms of the model size. LungSE-Net integrates a Squeeze-and-Excitation (SE) block into a lightweight CNN model for the purpose of image classification tasks. Purpose of integrating SE block as a component of network architecture is to improve model’s performance by explicitly modeling the inter-dependencies among the channels of its convolutional features. This is accomplished by dynamically readjusting feature responses for each channel, without requiring to considerably increase the computational resources. As a result, the network’s ability to represent information is improved. So, we can say that it has significant potential to aid pathologists for pathological diagnoses. To access the model’s performance, we used publicly available LC25000 dataset which consist of histopathological images, and obtained overall accuracy of 99.5%.