1
|
Tang YW, Ji J, Lin JW, Wang J, Wang Y, Liu Z, Hu Z, Yang JF, Ng TK, Zhang M, Pang CP, Cen LP. Automatic Detection of Peripheral Retinal Lesions From Ultrawide-Field Fundus Images Using Deep Learning. Asia Pac J Ophthalmol (Phila) 2023; 12:284-292. [PMID: 36912572 DOI: 10.1097/apo.0000000000000599] [Citation(s) in RCA: 4] [Impact Index Per Article: 4.0] [Reference Citation Analysis] [Abstract] [MESH Headings] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 09/20/2022] [Accepted: 12/19/2022] [Indexed: 03/14/2023] Open
Abstract
PURPOSE To establish a multilabel-based deep learning (DL) algorithm for automatic detection and categorization of clinically significant peripheral retinal lesions using ultrawide-field fundus images. METHODS A total of 5958 ultrawide-field fundus images from 3740 patients were randomly split into a training set, validation set, and test set. A multilabel classifier was developed to detect rhegmatogenous retinal detachment, cystic retinal tuft, lattice degeneration, and retinal breaks. Referral decision was automatically generated based on the results of each disease class. t -distributed stochastic neighbor embedding heatmaps were used to visualize the features extracted by the neural networks. Gradient-weighted class activation mapping and guided backpropagation heatmaps were generated to investigate the image locations for decision-making by the DL models. The performance of the classifier(s) was evaluated by sensitivity, specificity, accuracy, F 1 score, area under receiver operating characteristic curve (AUROC) with 95% CI, and area under the precision-recall curve. RESULTS In the test set, all categories achieved a sensitivity of 0.836-0.918, a specificity of 0.858-0.989, an accuracy of 0.854-0.977, an F 1 score of 0.400-0.931, an AUROC of 0.9205-0.9882, and an area under the precision-recall curve of 0.6723-0.9745. The referral decisions achieved an AUROC of 0.9758 (95% CI= 0.9648-0.9869). The multilabel classifier had significantly better performance in cystic retinal tuft detection than the binary classifier (AUROC= 0.9781 vs 0.6112, P < 0.001). The model showed comparable performance with human experts. CONCLUSIONS This new DL model of a multilabel classifier is capable of automatic, accurate, and early detection of clinically significant peripheral retinal lesions with various sample sizes. It can be applied in peripheral retinal screening in clinics.
Collapse
Affiliation(s)
- Yi-Wen Tang
- Joint Shantou International Eye Center of Shantou University and The Chinese University of Hong Kong, Shantou, Guangdong, China
| | - Jie Ji
- Network and Information Center, Shantou University, Shantou, Guangdong, China
| | - Jian-Wei Lin
- Joint Shantou International Eye Center of Shantou University and The Chinese University of Hong Kong, Shantou, Guangdong, China
| | - Ji Wang
- Joint Shantou International Eye Center of Shantou University and The Chinese University of Hong Kong, Shantou, Guangdong, China
| | - Yun Wang
- Joint Shantou International Eye Center of Shantou University and The Chinese University of Hong Kong, Shantou, Guangdong, China
| | - Zibo Liu
- Joint Shantou International Eye Center of Shantou University and The Chinese University of Hong Kong, Shantou, Guangdong, China
| | - Zhanchi Hu
- Joint Shantou International Eye Center of Shantou University and The Chinese University of Hong Kong, Shantou, Guangdong, China
| | - Jian-Feng Yang
- Joint Shantou International Eye Center of Shantou University and The Chinese University of Hong Kong, Shantou, Guangdong, China
| | - Tsz Kin Ng
- Joint Shantou International Eye Center of Shantou University and The Chinese University of Hong Kong, Shantou, Guangdong, China
- Shantou University Medical College, Shantou, Guangdong, China
- Department of Ophthalmology and Visual Sciences, The Chinese University of Hong Kong, Hong Kong
| | - Mingzhi Zhang
- Joint Shantou International Eye Center of Shantou University and The Chinese University of Hong Kong, Shantou, Guangdong, China
| | - Chi Pui Pang
- Joint Shantou International Eye Center of Shantou University and The Chinese University of Hong Kong, Shantou, Guangdong, China
- Department of Ophthalmology and Visual Sciences, The Chinese University of Hong Kong, Hong Kong
| | - Ling-Ping Cen
- Joint Shantou International Eye Center of Shantou University and The Chinese University of Hong Kong, Shantou, Guangdong, China
| |
Collapse
|
2
|
Ultra-wide-field fundus photography compared to ophthalmoscopy in diagnosing and classifying major retinal diseases. Sci Rep 2022; 12:19287. [PMID: 36369463 PMCID: PMC9650656 DOI: 10.1038/s41598-022-23170-4] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 04/07/2022] [Accepted: 10/26/2022] [Indexed: 11/13/2022] Open
Abstract
To analyze the performance of ultra-wide-field (UWF) fundus photography compared with ophthalmoscopy in identifying and classifying retinal diseases. Patients examined for presumed major retinal disorders were consecutively enrolled. Each patient underwent indirect ophthalmoscopic evaluation, with scleral depression and/or fundus biomicroscopy, when clinically indicated, and mydriatic UWF fundus imaging by means of CLARUS 500™ fundus camera. Each eye was classified by a clinical grader and two image graders in the following groups: normal retina, diabetic retinopathy, vascular abnormalities, macular degenerations and dystrophies, retinal and choroidal tumors, peripheral degenerative lesions and retinal detachment and myopic alterations. 7024 eyes of new patients were included. The inter-grader agreement for images classification was perfect (kappa = 0.998, 95% Confidence Interval (95%CI) = 0.997-0.999), as the two methods concordance for retinal diseases diagnosis (kappa = 0.997, 95%CI = 0.996-0.999) without statistically significant difference. UWF fundus imaging might be an alternative to ophthalmoscopy, since it allows to accurately classify major retinal diseases, widening the range of disorders possibly diagnosed with teleophthalmology. Although the clinician should be aware of the possibility that a minority of the most peripheral lesions may be not entirely visualized, it might be considered a first line diagnostic modality, in the context of a full ophthalmological examination.
Collapse
|
3
|
Russell MW, Muste JC, Rachitskaya AV, Talcott KE, Singh RP, Mammo DA. Visual, Anatomic Outcomes, and Natural History of Retinal Nerve Fiber Layer Schisis in Patients Undergoing Epiretinal Membrane Surgery. Ophthalmol Retina 2022; 7:325-332. [PMID: 36280203 DOI: 10.1016/j.oret.2022.10.011] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Track Full Text] [Journal Information] [Subscribe] [Scholar Register] [Received: 08/24/2022] [Revised: 10/10/2022] [Accepted: 10/17/2022] [Indexed: 11/05/2022]
Abstract
PURPOSE To evaluate the anatomic and visual outcomes of patients with idiopathic epiretinal membranes (ERMs) complicated by schisis of the retinal nerve fiber layer (sRNFL) in routine clinical practice. DESIGN Retrospective case-control study. PARTICIPANTS Patients undergoing idiopathic ERM surgery at Cole Eye Institute from 2013 to 2021. METHODS Patients were grouped by the presence or absence of sRNFL before surgery. Preoperative and postoperative data were collected regarding visual acuity (VA), changes in central subfield thickness (CST) over time, and presence of cystoid macular edema. MAIN OUTCOME MEASURES Frequency of sRNFL in patients undergoing idiopathic ERM surgery. RESULTS Overall, 48 (53.9%) of 89 patients presented with sRNFL. Schisis of the retinal nerve fiber layer patients presented with significantly decreased VA compared with those without (58.63 ± 12.48 vs. 67.68 ± 7.84 ETDRS letters, P < 0.001, respectively). At the final follow-up after ERM removal, there was no significant difference in final VA in patients with sRNFL compared with those without (71.16 ± 2.93 vs. 74.11 ± 2.76, P = 0.467). At presentation, patients with sRNFL had greater CST than those without (454 ± 10.01 vs. 436 ± 0.23, P = 0.23). This difference persisted at the 90-day follow-up after ERM removal (402 ± 8.08 vs. 375 ± 10.19 μm, P = 0.043). The resolution of sRNFL was reported at postoperative week 1 in 30 (96.7%) of 31 cases. CONCLUSIONS Schisis of the retinal nerve fiber layer is a microstructural feature in > 50% of idiopathic ERMs in routine clinical practice and carries visual significance on presentation and anatomic significance postoperatively. FINANCIAL DISCLOSURE(S) Proprietary or commercial disclosure may be found after the references.
Collapse
|