RePoint-Net detection and 3DSqU² Net segmentation for automatic identification of pulmonary nodules in computed tomography images

Abstract

Lung cancer is a leading cause of cancer-related deaths. Computer-aided detection (CAD) has emerged as a valuable tool to assist radiologists in the automated detection and segmentation of pulmonary nodules using Computed Tomography (CT) scans, indicating early stages of lung cancer. However, detecting small nodules remains challenging. This paper proposes novel techniques to address this challenge, achieving high sensitivity and low false-positive nodule identification using the RePoint-Net detection networks. Additionally, the 3DSqU2 Net, a novel nodule segmentation approach incorporating full-scale skip connections and deep supervision, is introduced. A 3DCNN model is employed for nodule candidate classification, generating final classification results by combining previous step outputs. Extensive training and testing on the LIDC/IDRI public lung CT database dataset validate the proposed model, demonstrating its superiority over human specialists with a remarkable 97.4% sensitivity in identifying nodule candidates. Moreover, CT texture analysis accurately differentiates between malignant and benign pulmonary nodules due to its ability to capture subtle tissue characteristic differences. This approach achieves a 95.8% sensitivity in nodule classification, promising non-invasive support for clinical decision-making in managing pulmonary nodules and improving patient outcomes.

Keywords

• Deep learning convolutional neural network
• computed tomography scans
• automated detection
• nodule segmentation
• nodule candidate classification