Implementation of 3D U-NET for Lung Cancer Classification

Cancer is a significant health issue that is spreading quickly worldwide and is challenging to control. Various types of cancer negatively impact people’s quality of life and considerably raise mortality rates. Factors such as smoking, genetic predisposition, and air pollution significantly contribute to the development of lung cancer. Lung cancer is one of the most common types of cancer worldwide and can progress rapidly, posing a serious threat to life if left untreated. However, if the disease is diagnosed early, the chances of successful treatment significantly increase, leading to better patient survival rates. Early diagnosis of lung cancer is critically essential for extending patient life spans. Research into early diagnosis and detection are examples of effective assistive treatment that continue to develop nowadays. For this reason, early diagnosis of lung cancer is of critical importance in terms of extending the life span of patients. Nowadays, studies on early diagnosis and correct treatment continue unabated. In this project, a deep learning model based on the 3D U-Net architecture was developed to detect lung cancer. The model assists in the earlier and more accurate diagnosis of abnormalities in medical images using deep learning techniques. The study utilized the "Lung-PET-CT-Dx" algorithms on a large dataset of 232,389 DICOM images for lung cancer diagnosis. Various data preprocessing techniques were implemented during the implementation phase, leading to a proposed model that demonstrated 95.64% accuracy in diagnosing lung cancer and staging and detecting medical abnormalities. We believe that the developed model will shed light on new studies on applying deep learning for the early detection of lung cancer.