Deep Learning-Based Lung Nodule Segmentation and Detection using CT Images

Abstract

Early identification of lung cancer significantly increases patient survival rates; however, accurate detection from Computed Tomography (CT) scans remains a challenging task due to variations in nodule size, shape, and intensity. This paper proposes an enhanced deep learning framework for lung nodule segmentation and detection. The approach combines an improved preprocessing pipeline, an attention-guided U-Net architecture, and a lightweight convolutional neural network (CNN) for classification. The preprocessing stage includes normalization and contrast enhancement to improve image quality. For segmentation, an attention-based U-Net is employed with a hybrid loss function that integrates Dice loss and binary cross-entropy to mitigate class imbalance. The segmented lung regions are subsequently analyzed using a compact CNN model for classification. Experimental evaluation demonstrates improved performance in terms of Dice coefficient, Intersection over Union (IoU), and classification accuracy when compared with baseline models. The proposed framework is computationally efficient and suitable for real-time clinical decision support systems.