Lung cancer detection based on machine learning

Lung cancer remains a leading cause of global cancer-related mortality. Early detection and accurate identification of lung nodules in computed tomography (CT) scans significantly improve prognosis but pose clinical challenges due to small lesion sizes, variability in nodule appearance, and overlapping anatomical structures. Conventional computer-aided detection methods have struggled with adaptability and accuracy. To address these issues, this paper introduces YOLOv11, a transformer-augmented deep learning architecture optimized for lung nodule detection. YOLOv11 integrates transformer blocks for enhanced global context modeling and convolutional block attention modules (CBAM) to prioritize crucial anatomical features. Experiments conducted on the LIDC-IDRI dataset indicate superior performance, achieving a mean average precision (mAP) of 86.4%, significantly outperforming baseline CNN models such as U-Net and TransUnet. Furthermore, YOLOv11 demonstrates robust real-time capabilities with inference speeds suitable for clinical deployment. This research underscores the potential of transformer-enhanced models to advance clinical diagnostics, improve early cancer detection, and ultimately reduce lung cancer mortality rates.