Development of a Lung Cancer Detection System Using Residual Network 50 on Computed Tomography Scans

Lung cancer has continued to be the primary cause of deaths from cancer worldwide, and this is due to the late diagnosis of this disease. This study shows how deep learning can be used to detect lung cancer at an early stage, using lung CT scan images. A pre-trained Residual Network with 50 layers (ResNet50) model was used with transfer learning to classify lung CT images as either cancerous or non-cancerous. The CT images used in this study were sourced from three publicly available datasets: the IQ-OTH/NCCD, the Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI), and the Bernard Institute of Radiology (BIR) Lung dataset. Since the original dataset had a majority of one class, it was balanced to a 50:50 ratio, resulting in a total of 1,720 images. The images were then split into training, validation, and testing sets at a 70:15:15 ratio. The model showed strong performance in testing, achieving an accuracy of 98.46%, a recall of 99.21%, a precision of 97.69%, and an F1-score of 98.45%. The trained model was then built into a web application, giving healthcare professionals a useful tool for classifying lung CT scans. In conclusion, this system has shown the role that deep learning can play in detecting lung cancer at an earlier stage and helping doctors make better and faster decisions.