Automated Classification of Lung Cancer from Histopathological Images using Deep Learning

Cancer is one of the leading causes of death worldwide, and its early detection plays a key role in effective management. The standard diagnostic method is biopsy and manual examination of histopathological slides by pathologists, which is time-consuming, requires high expertise, and is prone to error. Therefore, the use of computer-aided diagnostic systems based on artificial intelligence can help improve the efficiency of medical image interpretation. AI methods have multiplied the ability to extract quantitative information from digital histopathological images and are expected to reduce the workload for human experts while improving the accuracy and consistency of histopathological reports. In this research, various deep learning models, including ResNet50, ResNet152v2, and Inception-v3, were employed for lung cancer detection and subtype classification from histopathological images. The results of this study on the LC25000 dataset, utilizing transfer learning, demonstrate the significant potential of the proposed approach in diagnosing this disease. Among these, the ResNet152v2 network achieved the best performance, attaining 99.6% accuracy, 99.6% precision, and 99.6% sensitivity. However, after applying data leakage control and rigorous patient-wise separation of images into training and test sets, these values decreased to 96.9%, 96.7%, and 96.7%, respectively, providing a more realistic evaluation of the model. Overall, the findings indicate the high capacity of deep learning methods in classifying lung cancer subtypes. Nevertheless, the generalizability of the results to clinical settings requires validation on independent and diverse datasets, including whole-slide images, in future studies.