Deep Learning in Image Classification Using Modified ResNet50 and Texture Features for Lung Cancer Detection

According to WHO data, in 2020 there were 30,843 deaths from lung cancer in Indonesia. Some of the factors that lead to lung cancer are smoking, air pollution, and an unhealthy diet. Early detection with a lung CT scan is critical in finding treatment alternatives and increasing the patient's life expectancy. Screening and diagnosis by Pulmonologist takes a long time, but can be accelerated with computational techniques and reduce the cost.This study proposes a deep learning-based lung cancer detection method. Researchers compared the performance of ResNet50, ResNet50V2, and a modified model that combines ResNet50V2 with GLCM texture features. The models were simulated to classify lung CT scan images into normal, benign, or malignant. From the simulation results, it was found that the ResNet50V2 with GLCM achieved the highest accuracy at 82%, 82% sensitivity, and 90% specificity with processing each step for 71.32 seconds on 15% testing data.