Optimizing CT-Based Lung Nodule Segmentation Using DeepFuse U-Net: An Improved Deep Learning Approach

Introduction: One of the leading causes of morbidity and death globally is lung cancer. Often, in its first stage, lung cancer is asymptomatic. Growth of abnormalities in the lungs is called pulmonary nodules. Among the pulmonary nodules are both benign and malignant types. Early lung cancer identification can significantly benefit from the timely detection and segmentation of pulmonary nodules in lung computed tomography (CT) studies. The numerous forms and small size of the lung nodules make segmentation a challenging task. In this study, the DeepFuse U-Net model is proposed to detect lung nodules tumors, which assists clinicians in making accurate diagnoses. Methods: This study recommends a three-stage novel neural network, DeepFuse U-Net, based on UNet, integrating residual connections for improved feature representation, dense connections for feature reuse and gradient flow, and CLAHE pre-processing to enhance image contrast. Results: The proposed method achieved a Dice Similarity Coefficient (DSC) of 0.9497 and Intersection over Union (IoU) of 0.9012, a precision of 0.9672, a recall of 0.9067, and an F1 score of 0.9347, outperforming baseline methods consistently. Discussion: By integrating dense connectivity with residual connections, our proposed model enhances feature extraction, accuracy, and boundary delineation of lung nodules. Conclusion: Future work will explore external validation on different datasets, exploration of multimodal and 3D approaches, and generalizability to real-world deployment in clinical environments.