基于深度学习的人体腰椎MRI图像自动分割

Automatic segmentation of human lumbar spine MRI images based on deep learning

腰椎间盘病变是导致下背部疼痛的主要原因之一,而腰椎磁共振成像(MRI)图像在其诊断中发挥了关键作用。本研究引入了一种基于深度学习的自动分割方法,旨在增强椎间盘形态结构的识别和分割,从而减轻医疗专业人员手动分割所带来的不便和不一致性。我们采用了著名的分割网络Mask-Rcnn(Mask Region-based Convolutional Neural Network),该网络以其卓越的特征提取能力、出色的目标检测性能和精确的实例分割结果而闻名,因此成为最佳选择。通过利用PyTorch中的神经网络模型库,我们重构了数据集接口并微调了输出层参数,以更好地适应识别和分割腰椎间盘的任务。本研究使用了包含1545张腰椎MRI图像的公开数据集,每张图像都标注了椎间盘等结构。在对数据集进行预处理以保留有关椎间盘的标注后,我们随机选择了450张图像进行测试,其余用于训练。在经过20个训练周期后,我们实现了97.7%的平均精度和98.6%的平均召回率,96.9%的DICE系数。本研究强调了基于深度学习的自动分割方法在显著改善腰椎MRI图像中椎间盘的识别和分割方面的潜力。这种方法在临床应用中具有巨大前景,可能提高疾病诊断的准确性和效率,减轻了医疗专业人员的负担。

Lumbar intervertebral disc pathology constitutes a major contributor to lower back pain,with lumbar spine Magnetic Resonance Imaging(MRI)images playing a pivotal role in its diagnosis.This study introduces a deep learning-based automatic segmentation method aimed at enhancing the recognition and segmentation of intervertebral disc morphology,thereby mitigating the inconveniences and inconsistencies associated with manual segmentation by healthcare professionals.We employed the renowned segmentation network,Mask-Rcnn(Mask Region-based Convolutional Neural Network),recognized for its exceptional feature extraction capabilities,adept object detection performance,and precise instancesegmentation outcomes,rendering it an optimal choice.By leveraging the neural network model library in PyTorch,we revamped the dataset interface and fine-tuned output layer parameters to better align with the task of identifying and segmenting lumbar intervertebral discs.This study utilized a publicly available dataset comprising 1545 lumbar spine MRI images,each annotated for structures including intervertebral discs.After dataset preprocessing to retain annotations pertaining to intervertebral discs,we randomly selected 450 images for testing,with the remainder utilized for training.Following 20 training epochs,we achieved an average precision of 97.7%and an average recall of 98.6%,a DICE coefficient of 96.9%..This research underscores the substantial potential of a deep learning-based automatic segmentation method to markedly improve the recognition and segmentation of intervertebral discs in lumbar spine MRI images.This methodology holds promise for clinical applications,potentially enhancing the accuracy and efficiency of disease diagnosis while alleviating the burden on healthcare professionals.