Implementation of Deep Learning-Based Segmentation Technique on LDCT Scan Images for Detection of Lung Cancer in Early Stages

Early detection of lung cancer considerably increases the chances of survival and treatment efficacy. This paper studies the use of YOLOv8-based segmentation technique for automatic lung segment detection from LDCT scan images, with the goal of improving diagnostic accuracy and efficiency. The proposed model uses the YOLOv8 architecture, which is adapted for lung nodule identification and distinction in LDCT images through segmentation tasks. This technique makes use of YOLO's real-time object identification capabilities to swiftly and reliably identify troublesome spots. The model undergoes a thorough preparation procedure that includes image normalization, resampling, and augmentation in order to ensure robustness and generalizability. A novel loss function is used for training the model, combining metrics specific to segmentation with traditional object identification loss components. The model's performance is evaluated using metrics such as F1 Score, Recall, Precision, ROC-AUC and Intersection over Union (IoU), which demonstrate how well the model segments lung nodules. According to the study's findings, the YOLOv8-based segmentation method presents a potentially helpful tool for identifying lung cancer in early stages. This might lessen radiologist's workload and enhance patient outcomes by enabling early diagnosis and treatment.