Evaluation of multi-classification method of color fundus photograph quality based on ResNet50-OC / 中华实验眼科杂志

Objective:To evaluate the efficiency of ResNet50-OC model based on deep learning for multiple classification of color fundus photographs.Methods:The proprietary dataset (PD) collected in July 2018 in BenQ Hospital of Nanjing Medical University and EyePACS dataset were included.The included images were classified into five types of high quality, underexposure, overexposure, blurred edges and lens flare according to clinical ophthalmologists.There were 1 000 images (800 from EyePACS and 200 from PD) for each type in the training dataset and 500 images (400 from EyePACS and 100 from PD) for each type in the testing dataset.There were 5 000 images in the training dataset and 2 500 images in the testing dataset.All images were normalized and augmented.The transfer learning method was used to initialize the parameters of the network model, on the basis of which the current mainstream deep learning classification networks (VGG, Inception-resnet-v2, ResNet, DenseNet) were compared.The optimal network ResNet50 with best accuracy and Micro F1 value was selected as the main network of the classification model in this study.In the training process, the One-Cycle strategy was introduced to accelerate the model convergence speed to obtain the optimal model ResNet50-OC.ResNet50-OC was applied to multi-class classification of fundus image quality.The accuracy and Micro F1 value of multi-classification of color fundus photographs by ResNet50 and ResNet50-OC were evaluated.Results:The multi-classification accuracy and Micro F1 values of color fundus photographs of ResNet50 were significantly higher than those of VGG, Inception-resnet-v2, ResNet34 and DenseNet.The accuracy of multi-classification of fundus photographs in the ResNet50-OC model was 98.77% after 15 rounds of training, which was higher than 98.76% of the ResNet50 model after 50 rounds of training.The Micro F1 value of multi-classification of retinal images in ResNet50-OC model was 98.78% after 15 rounds of training, which was the same as that of ResNet50 model after 50 rounds of training.Conclusions:The proposed ResNet50-OC model can be accurate and effective in the multi-classification of color fundus photograph quality.One-Cycle strategy can reduce the frequency of training and improve the classification efficiency.

Clinical validation of RIA-G, an automated optic nerve head analysis software

Purpose: To clinically validate a new automated glaucoma diagnosis software RIA-G. Methods: A double-blinded study was conducted where 229 valid random fundus images were evaluated independently by RIA-G and three expert ophthalmologists. Optic nerve head parameters [vertical and horizontal cup–disc ratio (CDR) and neuroretinal rim (NRR) changes] were quantified. Disc damage likelihood scale (DDLS) staging and presence of glaucoma were noted. The software output was compared with consensus values of ophthalmologists. Results: Mean difference between the vertical CDR output by RIA-G and the ophthalmologists was ? 0.004 ± 0.1. Good agreement and strong correlation existed between the two [interclass correlation coefficient (ICC) 0.79; r = 0.77, P < 0.005]. Mean difference for horizontal CDR was ? 0.07 ± 0.13 with a moderate to strong agreement and correlation (ICC 0.48; r = 0.61, P < 0.05). Experts and RIA-G found a violation of the inferior–superior NRR in 47 and 54 images, respectively (Cohen's kappa = 0.56 ± 0.07). RIA-G accurately detected DDLS in 66.2% cases, while in 93.8% cases, output was within ± 1 stage (ICC 0.51). Sensitivity and specificity of RIA-G to diagnose glaucomatous neuropathy were 82.3% and 91.8%, respectively. Overall agreement between RIA-G and experts for glaucoma diagnosis was good (Cohen's kappa = 0.62 ± 0.07). Overall accuracy of RIA-G to detect glaucomatous neuropathy was 90.3%. A detection error rate of 5% was noted. Conclusion: RIA-G showed good agreement with the experts and proved to be a reliable software for detecting glaucomatous optic neuropathy. The ability to quantify optic nerve head parameters from simple fundus photographs will prove particularly useful in glaucoma screening, where no direct patient–doctor contact is established.

Diagnostic Availability of Blind Spot Mapping for Ocular Torsion

PURPOSE: To evaluate diagnostic the usefulness of blind spot mapping in measuring ocular torsion changes and to investigate the correlations of inferior oblique muscle overaction (IOOA) and excyclotorsion measurements using fundus photographs and blind spot mapping in patients with secondary IOOA. METHODS: Eleven patients (12 eyes; IOOA group) diagnosed with secondary IOOA were evaluated for ocular movement, fundus photograph and Humphrey standard automated perimetry, and 10 patients (20 eyes; control group) were subjected to the same tests. An ocular movement examination was performed to evaluate IOOA, and fundus photograph and Humphrey standard automated perimetry were used to measure the ocular torsion. Inferior oblique myectomy or recession was performed along with horizontal strabismus surgery, and preoperative and postoperative IOOA and ocular torsion measurements were compared between the groups. RESULTS: In the IOOA group after surgery, the IOOA decreased from +2.42 ± 0.63 to +0.50 ± 0.52, the ocular torsion decreased from +14.15 ± 3.60° to +7.47 ± 1.65° (p < 0.001) on fundus photographs, and from +12.19 ± 1.62° to +9.69 ± 1.75° (p = 0.061) in Humphrey standard automated perimetry. The control group showed a mean ocular torsion of 7.44 ± 1.62° on fundus photographs and +7.24 ± 1.28° on Humphrey standard automated perimetry. CONCLUSIONS: The usefulness of blind spot mapping when the ocular torsion was measured in IOOA patients was considered low, due to the weak correlation between IOOA and extorsion; preoperative and postoperative ocular torsion amount values were not significantly different.

Analysis of Localized Retinal Nerve Fiber Layer Defects not Detected by Optical Coherence Tomography

PURPOSE: To analyze localized RNFL defect cases that were identified in retinal nerve fiber layer (RNFL) fundus photographs but not in optical coherence tomography (OCT). METHODS: Analysis of OCT scans and images was performed for 14 eyes (17 locations) that showed localized RNFL defects in RNFL fundus photographs but not in RNFL thickness average analysis. RESULTS: With respect to the range of RNFL defects, 41.2% were less than 10degrees, 47.0% were 11 to 20degrees, and 11.8% were 21 to 30degrees. In 71.4% of the RNFL cases the defects were less than 10degrees and the decrease of RNFL thickness was not readily observable on the OCT scan images. In all cases of RNFL defects in the 11 to 30degrees range the decrease in RNFL thickness could be assessed on the OCT scan images. Nonetheless, the decrease of RNFL thickness could not be seen on the OCT analysis images in which the results of the RNFL thickness made through an automated computer algorithm were displayed. CONCLUSIONS: The range of localized RNFL defects that were difficult to detect with OCT consisted of those cases that were almost less than 20degrees. The limitations of the OCT scan itself in patients with RNFL with an angular width defect less than 10degrees and the problems of RNFL thickness analysis processing in patients with an angular width of 11 to 30degrees may decrease the sensitivity of OCT in diagnosing RNFL defects.

The Abnormalities of Retinal Arteriole in Atherothrombotic Ischemic Stroke Patients Representing the Changes of Cerebral Vasculature Indirectly

PURPOSE: Retinal blood vessels and cerebral small vessels possess similar characteristics anatomically, physiologically and embryologically. We studied the availability of abnormal fundus findings of patients who had an atherothrombotic ischemic stroke and who have the risk factors. METHODS: Fundus photographs and brain images were taken in patients who had a first-ever symptomatic ischemic stroke of large artery atherosclerosis (LAA) or small vessel occlusion (SVO) from March 2004 to February 2005. We analyzed the association between fundus abnormalities and ischemic stroke subtypes. RESULTS: Based on brain MRI and MRA, a total of 47 patients were classified into SVO and LAA groups. The SVO group consisted of 27 patients (mean age: 69.7 years), and the LAA group consisted of 20 patients (mean age: 65.4 years). The control group comprised 15 patients (mean age: 64.9 years). The baseline characteristics were similar among the three groups. The severity of the retinal arteriolar narrowing and sclerosis were associated with hypertension. Compared to the control group, both the SVO and LAA groups showed more severe arteriolar sclerosis, the SVO group showed more severe arteriolar narrowing and the LAA group showed more frequent AV crossing and retinal exudate. CONCLUSIONS: Retinal arteriolar abnormalities such as arteriolar narrowing and sclerosis are more severe in atherothrombotic ischemic stroke patients. Indirectly, retinal microvascular changes may indicate the status of the cerebral vasculature. Thus, analysis of fundus findings is useful for predicting an atherothrombotic ischemic stroke and planning follow-up examinations.

The Detection of Retinal Never Fiber Layer Defect by Modification of Non-mydriatic Digital Fundus Photograph

PURPOSE: To evaluate the efficiency of digital retinal nerve fiber layer (RNFL) photographs converted from a non-mydriatic digital fundus camera for detecting RNFL defects. METHODS: Ninety-five eyes were evaluated with both a non-mydriatic digital fundus photograph and a digital RNFL photograph by two glaucoma specialists independently. The red-free, monochrome modified digital fundus photograph was acquired from non-mydriatic digital fundus photographs using Photoshop 7.0. Whether a localized wedge-shaped or diffuse RNFL defect existed or not was evaluated on a non-mydriatic digital fundus photograph and a modified digital fundus photograph, and inter- and intraobserver agreement were also evaluated. Regarding the use of digital RNFL photographs as a standard method, we calculated the sensitivity, specificity, and positive and negative predictive values of the two types of photographs. RESULTS Interobserver agreement (Cohen's kappa values) about localized RNFL defects observable on digital RNFL photographs, non-mydriatic digital fundus photographs, and modified digital fundus photographs were 0.749, 0.634, and 0.793, respectively, but all were 0.417 or less with regard to diffuse RNFL defects. Regarding localized RNFL defects, the sensitivity, specificity, and positive and negative predictive values of modified digital fundus photographs were 85.7%, 95.5%, 85.7% and 94.1%, respectively, showing superiority to those of non-mydriatic digital fundus photographs. But, in cases of diffuse RNFL defect, the sensitivity and specificity of modified digital fundus photographs were 60.0% and 97.8%, respectively, indicating no superiority to non-mydriatic digital fundus photographs. CONCLUSIONS: The localized RNFL defect can be detected more efficiently with modified digital fundus photographs than non-mydriatic digital fundus photographs, but the diffuse RNFL defect cannot.

The Detection of Retinal Never Fiber Layer Defect by Modification of Non-mydriatic Digital Fundus Photograph

PURPOSE: To evaluate the efficiency of digital retinal nerve fiber layer (RNFL) photographs converted from a non-mydriatic digital fundus camera for detecting RNFL defects. METHODS: Ninety-five eyes were evaluated with both a non-mydriatic digital fundus photograph and a digital RNFL photograph by two glaucoma specialists independently. The red-free, monochrome modified digital fundus photograph was acquired from non-mydriatic digital fundus photographs using Photoshop 7.0. Whether a localized wedge-shaped or diffuse RNFL defect existed or not was evaluated on a non-mydriatic digital fundus photograph and a modified digital fundus photograph, and inter- and intraobserver agreement were also evaluated. Regarding the use of digital RNFL photographs as a standard method, we calculated the sensitivity, specificity, and positive and negative predictive values of the two types of photographs. RESULTS Interobserver agreement (Cohen's kappa values) about localized RNFL defects observable on digital RNFL photographs, non-mydriatic digital fundus photographs, and modified digital fundus photographs were 0.749, 0.634, and 0.793, respectively, but all were 0.417 or less with regard to diffuse RNFL defects. Regarding localized RNFL defects, the sensitivity, specificity, and positive and negative predictive values of modified digital fundus photographs were 85.7%, 95.5%, 85.7% and 94.1%, respectively, showing superiority to those of non-mydriatic digital fundus photographs. But, in cases of diffuse RNFL defect, the sensitivity and specificity of modified digital fundus photographs were 60.0% and 97.8%, respectively, indicating no superiority to non-mydriatic digital fundus photographs. CONCLUSIONS: The localized RNFL defect can be detected more efficiently with modified digital fundus photographs than non-mydriatic digital fundus photographs, but the diffuse RNFL defect cannot.