Development and application of an accurate retinal vascular network segmentation method for multiple diseases based on a multi-path network
10.3760/cma.j.cn115989-20240731-00215
- VernacularTitle:基于多路径网络的多病种精准视网膜血管网络分割方法的建立及应用
- Author:
Jinze ZHANG
1
;
Jiaxiong LI
;
Gengyuan WANG
;
Jin YUAN
;
Peng XIAO
Author Information
1. 中山大学中山眼科中心 眼病防治全国重点实验室 广东省眼科视觉科学重点实验室，广州 510060
- Keywords:
Retinal diseases;
Retinal vessels;
Artificial intelligence;
Color fundus photography;
Morphological parameters
- From:
Chinese Journal of Experimental Ophthalmology
2024;42(12):1120-1126
- CountryChina
- Language:Chinese
-
Abstract:
Objective:To establish an accurate retinal vascular network segmentation method for multiple fundus diseases, and to investigate the changing patterns of retinal vascular morphological parameters in these diseases.Methods:A retrospective study was conducted.Color fundus photography data of 829 patients with fundus diseases and 146 healthy adults were collected at Zhongshan Ophthalmic Center, Sun Yat-sen University from January 2020 to December 2023.The multi-path segmentation network was fine-tuned, and the color fundus photography data of diabetic retinopathy (DR), glaucoma and age-related macular degeneration (AMD) patients and healthy adults in the fundus image vessel segmentation public dataset were input for training until the loss value of the model stopped decreasing, and finally the trained multi-disease retinal vascular segmentation model was obtained.The retinal blood vessel morphological characteristics analysis method previously developed by our research group was used to analyze the subjects' color fundus images centered on the macula, the retinal blood vessel fractal dimension (D f), vascular area ratio (VAR), mean diameter (D m), tortuosity (τ) and other morphological characteristics parameters were extracted and compared among various disease groups.This study adhered to the Declaration of Helsinki.The study protocol was approved by the Ethics Committee of Zhongshan Ophthalmic Center, Sun Yat-sen University (No.2023KYPJ344).Written informed consent was obtained from each subject. Results:The accuracy of the multi-disease color fundus photography vessel segmentation model on the test set was 0.987, and the area under the receiver operating characteristic curve was 0.995.After adjustment for age and sex, there were statistically significant differences in adjusted D f, adjusted VAR, adjusted D m and τ among different groups ( F=27.87, 47.60, 26.48, 4.63; all at P<0.001).Adjusted D f in AMD group, DR group, diabetic macular edema (DME) group, retinitis pigmentosa (RP) group, branch retinal vein occlusion (BRVO) group and central retinal vein occlusion (CRVO) group was significantly decreased than in normal control group, and the differences were statistically significant (all at P<0.05).Adjusted VAR in all disease groups except optic neuritis group and central serous chorioretinopathy group was significantly decreased compared with normal control group, and the differences were statistically significant (all at P<0.05).The adjusted D m in DME, glaucoma, RP, BRVO and CRVO groups was significantly decreased than that in normal control group, and the differences were statistically significant (all at P<0.05).τ was not affected by age or sex and did not require adjustment.τ in DR group and DME group was significantly increased compared with normal control group, and the differences were statistically significant (both at P<0.05). Conclusions:An accurate retinal blood vessel segmentation method for various fundus diseases was successfully constructed.This method shows high accuracy in retinal blood vessel segmentation in color fundus photographs of various retinal diseases.There are significant differences in the morphological characteristics of retinal blood vessels among different retinal diseases.
