A comprehensive review of deep learning, traditional methods, and GenAI techniques for lung nodule segmentation

This paper offers an in-depth analysis of advanced techniques in segmenting lung nodules, a critical aspect in the early detection and diagnosis of lung cancer. The analysis includes a wide spectrum of methods, ranging from traditional image processing techniques to cutting-edge deep learning approaches, thus providing a comprehensive overview of the current state-of-the-art in lung nodule segmentation. Among the deep learning techniques, in this review, the intricacies of nested 3D fully connected convolutional networks, Wavelet U-Net++, SKV-Net, SAtUNet, Dense Atrous-Net, U-Det, DEHA-Net, and CoLe-Convolutional Neural Network were explored, with their architectures and performance metrics highlighted. Additionally, traditional image processing methods such as the Fast-Marching Method and Fast Multi-Crop Guided Attention were examined. Furthermore, generative AI approaches including Segment Anything Model and Segment Everything Everywhere All at Once were introduced to the Lung Image Database Consortium and Image Database Resource Initiative dataset, which contains comprehensive nodule annotations, and were used to test these approaches on a subset of 100 pictures. Their results were compared against the methods above. By examining the strengths and limitations of each approach, insights into the current landscape of lung nodule segmentation methods were provided, along with directions for future research.