AutoLungDx: A hybrid deep learning approach for early lung cancer diagnosis using 3D Res-U-Net, YOLOv5, and vision transformers

Objective: Lung cancer is a leading cause of cancer-related death worldwide, and early detection is crucial to improving patient outcomes. However, early diagnosis is a major challenge, particularly in low-resource settings with limited access to CT resources and trained radiologists. This study aims to propose an automated end-to-end deep learning-based framework for the early detection and classification of lung nodules, specifically for low-resource settings. Methods: The proposed framework consists of three stages: lung segmentation using the proposed 3D Res-U-Net, nodule detection using YOLO-v5, and classification with a Vision Transformer-based architecture. We evaluated the proposed framework on the publicly available dataset, LUNA16. The performance of the proposed framework was evaluated using task-specific metrics. Results: The proposed framework achieved a 98.82% lung segmentation dice score while achieving 0.76 mAP@50 for nodule detection from the segmented lung at a low false positive rate. Furthermore, our proposed Vision Transformer network achieved an accuracy of 96.29%, which is 4.25% higher than the state-of-the-art networks. The performance of all three networks in the proposed framework was compared with state-of-the-art studies, and they were found to outperform them across the reported metrics. Conclusion: Our proposed end-to-end deep learning-based framework can effectively segment the lung and detect and classify lung nodules, particularly in low-resource settings with limited access to radiologists. The proposed framework outperforms existing studies across all respective evaluation metrics. Significance: The proposed framework can potentially improve the accuracy and efficiency of lung cancer screening, ultimately leading to better patient outcomes, even in settings with limited medical resources and trained personnel.