ABSTRACT
Objective To evaluate the application of the standard manual labeling on identification of retinopathy of prematurity ( ROP) images in deep learning. Methods According to the International Classification of ROP,different periods of ROP were classified into stage disease and plus disease in this study. From Joint Shantou International Eye Center from August 2009 to July 2018, a total of 1464 labeled fundus retinal photographs were divided randomly by stratified sampling into 3 groups:stage disease group(subgroup 1:173,subgroup 2:117) was used to train for labeling stage disease,whereas plus disease group(subgroup 1:163,subgroup 2:116) was used to train for labeling plus disease,and consistent labels group consisted of 895 consistent labeled images on both disease. Graders consisted of senior experts,3 senior ophthalmologists and 2 interns,and received training for classification and labeling on ROP fundus images. The results were compared among the doctors and doctors with deep learning,and the agreement between non-experts doctors and the reference standards, and deep learning and the reference standards were tested. Results After the first training,the overall agreement rate of the senior ophthalmologist group and the intern group were lower than 90％ for both two disease labeling. After two to three times of training, in image of consistent labels group,overall agreement rates of senior ophthalmologists and intern doctor's were 98. 99％ ( Kappa=0. 979),99. 22％ (Kappa=0. 984) on stage disease,and 97. 43％ (Kappa=0. 914),98. 11％ (Kappa=0. 935) on plus disease,respectively. The agreement on stage disease using deep learning based on human-machine combination was 94. 08％,Kappa value was 0. 880,which achieved good degree. Conclusions Standardized manual labeling can improve the intelligentization of deep learning on identification of ROP images,and be considered as an innovative method of homogenization and standardized training for doctors in ophthalmology.
ABSTRACT
Background As myopia is a common ocular condition which has been reported as the risk factor of primary open angle glaucoma,it is of great importance to evaluate the optic disc morphology in myopic eyes.Objective This study was to evaluate the agreement of optic disc measurements between Cirrus high-density optical coherence tomography (HD-OCT) and Heidelberg retina tomograph (HRT) in myopic eyes;and to investigate the relationships between axial length (AL) and differences of optic disc parameters measured with the two devices.Methods One hundred and fifty myopic subjects were included in this prospective cross-sectional study.One eye from each subject was randomly selected for optic disc imaging by Cirrus HD-OCT and HRT2 in Joint Shantou International Eye Center of Shantou University and The Chinese University of Hong Kong from September to December in 2010 under the approval of Ethic Committee of this hospital and informed consent of each patient was received.Each subject received complete ophthalmic examinations including intraocular pressure (IOP) measurement,visual acuity,refraction,slit lamp,dilated fundus examination and perimetry.The subjects were divided into low (≤-3.00 D,35 eyes),moderate (-3.00 D＜SE＜-6.00 D,60 eyes) and high myopia (SE ≥-6.00 D,55 eyes) groups according to the refractive status.Measurement of axial length was performed with IOL master.Optic disc parameters including disc area,rim area,cup volume,vertical cup-to-disc ratio (VCDR) and average cup-to-disc area ratio (ACDR) were measured by Cirrus HD-OCT and HRT2,respectively.The OCT measurements were corrected for ocular magnification using the Littman's formula,and the results were compared between the instruments.The measurement agreement was evaluated using Bland-Altman plots.Pearson correlation analysis was used to evaluate the associations between AL and the measurement differences of the two instruments.Results The mean axial length and refraction were (25.62±1.10) mm and (-5.22±2.34) D,respectively.The corrected optic disc parameters were significantly larger than those without adjustment by using Cirrus HD-OCT (all at P＜ 0.001).In the high myopic group,the disc area measured by Cirrus HD-OCT was significantly larger than that by HRT2 (P＜0.001).In the moderate myopic group,the rim area measured by HRT2 was significantly larger than that by the Cirrus HD-OCT (P =0.040).The measurements of ACDR,VCDR and cup volume by Cirrus HD-OCT were all larger than those by HRT2 in the three myopic groups (all at P＜0.001).The 95％ limits of agreement (LoA) of disc area and rim area with the two devices were-0.64 to 0.74 and-0.74 mm to 0.62 mm2,respectively.The differences of disc area,rim area and cup volume measurements from the two devices were significantly and positively associated with axial length (r=0.158,0.148,0.156,all at P＜0.05).No significant correlation was detected between AL and the differences of ACDR and VCDR (r =0.012,0.093,both at P ＞ 0.05).Conclusions Optic disc parameters measured by Cirrus HD-OCT are affected by optical magnification in myopic eyes.Poor agreement is found across all of the disc measurements with Cirrus HD-OCT and HRT2.The two devices should not be used interchangeably for measurements of optic disc.Moreover,the differences between measurements of the two devices are significantly associated with AL.