A Neural Network Model in Identifying Non-Small-Cell Lung Cancer through CT Scans

Lung cancer is the leading cause of cancer-related deaths worldwide, causing approximately 1.8 million deaths in 2022 alone. Lung cancer is often detected through the use of Low-Dose Computed Tomography (LDCT) scans, which use small amounts of radiation to construct detailed pictures of regions in the body. This study aims to explore the ability of Artificial Intelligence, particularly Convolutional Neural Networks (CNNs), to detect lung cancer from CT scans. Using an online dataset consisting of 1000 images, a CNN was developed with ResNet50 as the base model used for feature extraction. The model achieved a validation accuracy of 98.78% and a testing accuracy of 97.53%. This showcases the proficiency of the model in detecting lung cancer. However, this was only when a binary classification system was implemented, where the model was made to simply determine the presence of cancer. The model faced great difficulty in distinguishing between the types of lung cancer: Adenocarcinoma, Squamous Cell Carcinoma, and Large Cell Carcinoma. Additionally, the presence of a small number of false negatives while testing shows the danger of relying on AI and demonstrates the necessity of further fine-tuning before practical use.