基于DeeplabV3+网络的睑板腺图像分割研究和评价

Research and evaluation of tarsal gland image segmentation based on DeeplabV3+network

目的:构建基于DeeplabV3+网络的人工智能(AI)系统算法模型,提高眼干燥症诊疗效率。方法:收集某医院干眼门诊就诊患者的睑板腺图像,构建图像数据库,随机分配为训练集和验证集,投入模型训练,分析并验证其可行性和有效性。结果:在内部验证集,基于DeeplabV3+的算法模型对于睑板腺区域分割的准确率达95.65%,均交并比和Kappa系数分别为83.75%和92.96%。该算法分割出的萎缩区域,与临床医生分割结果相似。结论:DeeplabV3+网络模型能够实现眼干燥症患者睑板腺腺体的自动切分,可辅助相关疾病的临床诊断和筛查,提高诊断效率。

Objective To construct an artificial intelligence(AI)system algorithm model based on DeeplabV3+network to improve the efficiency of ophthalmoxerosis diagnosis and treatment.Methods The tarsal gland images of ophthalmoxerosis patients in a hospital were collected and the image database was constructed.The images were randomly divided into a training set and a validation set,and the model trained to analyze and verify its feasibility and effectiveness.Results For the internal validation set,the accuracy of the algorithm model based on DeeplabV3+for tarsal gland region segmentation reached 95.65%,with the mIOU and Kappa coefficient of 83.75%and 92.96%,respectively.The atrophic regions segmented by this algorithm were similar to those segmented by clinical doctors.Conclusion The DeeplabV3+network model can achieve automatic segmentation of tarsal gland,which can assist clinical diagnosis and screening of related diseases and improve diagnostic efficiency.