深度学习在脊椎图像分割中的研究进展

Research Progress of Deep Learning in Spine Image Segmentation

影像学检查是临床诊断和治疗脊椎疾病的重要依据,脊椎的结构复杂,周围布满肌肉、血管和神经组织,且由于病理变化容易导致脊椎形状发生改变,对脊椎图像的精确分割能够辅助医生精准定位病灶,高效评估病变,引导手术治疗等。近年来,深度学习凭借其强大的特征学习能力在脊椎分割领域得到广泛的应用。为研究深度学习在脊椎图像分割任务中的研究现状和发展,对深度学习的经典框架进行比较,将脊椎图像分割中常用的数据集与评价指标归纳介绍,总结CNN、FCN、U-Net和GAN不同网络模型在脊椎分割中应用的最新进展,对模型在脊椎分割上的特点进行分析,详细讨论了当前深度学习在脊椎分割中遇到的问题和挑战,并对今后的发展方向做出展望。

Imaging is an important basis for clinical diagnosis and treatment of spinal diseases.The spine has a complex structure,surrounded by muscles,blood vessels and nerve tissues,and the shape of the spine is easily changed due to path-ological changes,so accurate segmentation of spine images can assist doctors to accurately locate lesions,efficiently eval-uate lesions and guide surgical treatment.In recent years,deep learning has been widely used in the field of spine segmen-tation with its powerful feature learning capability.To study the current research status and development of deep learning in spine image segmentation tasks,this paper compares the classical frameworks of deep learning,organizes the commonly used datasets and evaluation metrics in spine image segmentation,summarizes the latest progress of different network models of CNN,FCN,U-Net and GAN applied in spine segmentation,analyzes the characteristics of the models on spine segmentation,and discusses in detail the current deep learning in spine segmentation,discusses in detail the current prob-lems and challenges encountered in deep learning in spine segmentation,and presents an outlook on feasible future devel-opment directions.