Enhanced YOLO-Based Lung Cancer Detection Using Attention Mechanism and Multi-Scale Feature Fusion
Abstract
Lung cancer happens to be one of the most significant causes of deaths throughout the globe, necessitating efficient diagnosis for patient survival and accurate treatment plans. Although deep learning approaches continue to develop, there is still scope for bettering the efficiency in feature extraction, lesion localization, and multiclass detection. Therefore, in this study, we present a deep learning model-based lung cancer detection algorithm on the basis of a hybrid Conv, C3k2, SPPF, Upsample and Concat feature fusion blocks, C2PSA attention, and effective detection heads. The technique focuses on improving multi-level feature extraction, obtaining lesion-centered attention mechanism, and conducting multiclass localization of Normal, Benign Lung Nodule, and Malignant Lung Nodule classes. Results show that our suggested algorithm exceeds several advanced recent approaches concerning attention mechanism. Our technique achieved Precision of 0.847, Recall of 0.847, mAP of 0.909, and F1 Score of 0.847, recording the highest value of mAP among all evaluated approaches. It can be deduced that the technique can provide an efficient solution to lung cancer detection.