An Efficient Hybrid Framework U-Net-based deep learning Technique of Early Detection of Pulmonary Nodules using LIDC-IDRI Dataset

Abstract

Radiologists still face difficulties and mistakes while screening lung computed tomography (CT) images for pulmonary nodules, particularly small and inconspicuous malignant lesions. Frequent radiation exposure, the complexity of radiomic characteristics in low-dose CT scans, and the high cost of imaging therapy are some of the challenges. Our technique proposed a novel automated computer-aided diagnostic technique to address these issues by increasing the accuracy of early lung nodule diagnosis. We suggested a four-step technique that involves (1) a preprocessing step consisting of contrast-limited adaptive histogram equalization to refine the contrast of contribution inputs, followed by extracting and combining texture and shape features in parallel using a gray level co-occurrence matrix for the first features and region of interest (ROI) properties for the second. In addition, we suggested a hybrid U-Net-based deep learning architecture for categorization that successfully blends automatically learned features with manually created features. This integration improves the precision and resilience of pulmonary nodule classification by utilizing the convolutional neural networks (CNN)’s capacity to capture spatial hierarchies. We implemented our proposed technique on the overtly accessible Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) dataset, employing the Python programming language for the implementation. Experimental results confirmed that our technique attained segmentation and classification results of 95.35% accuracy, 95.33% sensitivity, 94.23% specificity, and 95.44% AUC rates, outperforming several state-of-the-art methods. This high-performance approach offers a reliable solution for early detection, potentially reducing lung cancer mortality rates through timely diagnosis.