基于深度学习的颈椎间盘突出识别方法

Recognition of Cervical Disc Herniation Based on Deep Learning

颈椎间盘突出在人群特别是老年人群中非常普遍,且具有多种类型,对其进行快速、有效且准确的识别并能检测其类型,具有重要意义。构建基于深度学习的颈椎间盘突出识别方法,其核心思想在于:在大量采集各种类型颈椎间盘突出患者颈部MRI影像样本的基础上,将样本分为膨出型、突出型、未患病三类,通过筛选、截取、归一化、扩增处理等操作,构建有效的训练数据;依据本数据集的特点,提出了一种改进的迁移学习网络模型(GoogleNet Inception V3)结构,将模型中的平均池化以及最大池化替换成了组合池化,并通过对数据样本进行训练,使得最终训练好的分类模型最优。通过大量测试数据进行测试,并采用了一种加权F1score的评估方式,结果表明:在召回率、精确度方面,针对本数据集,本模型都有显著优势。

Cervical disc herniation is very common in the population,especially the elderly population,and has a variety of types.Therefore,it is of great significance to quickly,effectively and accurately identify and detect its types.This paper build a recognition method based on the deep learning of cervical intervertebral disc herniation,its core idea is:in a large collection of various types of cervical intervertebral disc herniation in patients with cervical MRI image samples,on the basis of the sample is divided into bulging,prominent,not three,through screening,interception,normalization,amplification processing operations,such as building effective training data;According to the characteristics of this data set,an improved migration learning network model(GoogleNet Inception V3)structure is proposed.Average pooling and maximum pooling in the model are replaced by combined pooling.By training the data samples,the final trained classification model is optimized.Through a large number of test data,and using a weighted F1score evaluation method,the results show that:in terms of recall rate and accuracy,this model has significant advantages for this data set.