Yu-Chieh Ko
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Yu-Chieh Ko has a total of 57 co-authors. Here are the top co-authors by number of shared publications:
- Catherine Jui-Ling Liu
Department of Ophthalmology, Taipei Veterans General Hospital, Taipei 11217, Taiwan
6 shared publications Yu-Fan Chang
Institute of Clinical Medicine, National Yang Ming Chiao Tung University, Taipei, 112, Taiwan
6 shared publicationsCatherine Jui-Ling Liu
National Yang Ming Chiao Tung University School of Medicine, Taipei, Taiwan
5 shared publicationsShih-Hwa Chiou
Institute of Clinical Medicine, National Yang Ming Chiao Tung University, Taipei, 112, Taiwan
4 shared publicationsMei-Ju Chen
Department of Ophthalmology, Taipei Veterans General Hospital, Taipei, Taiwan, ROC
3 shared publications- Mei-Ju Chen
Department of Ophthalmology, Taipei Veterans General Hospital, Taipei, Taiwan, ROC
2 shared publications
Latest Publications
- 1 May 2023
- Published by Wolters Kluwer Health in Retina
- Vol. 43  (5) , 767-774
PREDICTING VISUAL OUTCOME AFTER SURGERY IN PATIENTS WITH IDIOPATHIC EPIRETINAL MEMBRANE USING A NOVEL CONVOLUTIONAL NEURAL NETWORK
Purpose: To develop a deep convolutional neural network that enables the prediction of postoperative visual outcomes after epiretinal membrane surgery based on preoperative optical coherence tomography images and clinical parameters to refine surgical decision making. Methods: A total of 529 patients with idiopathic epiretinal membrane who underwent standard vitrectomy with epiretinal membrane peeling surgery by two surgeons between January 1, 2014, and June 1, 2020, were enrolled. The newly developed Heterogeneous Data Fusion Net was introduced to predict postoperative visual acuity outcomes (improvement ≥2 lines in Snellen chart) 12 months after surgery based on preoperative cross-sectional optical coherence tomography images and clinical factors, including age, sex, and preoperative visual acuity. The predictive accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve of the convolutional neural network model were evaluated. Results: The developed model demonstrated an overall accuracy for visual outcome prediction of 88.68% (95% CI, 79.0%–95.7%) with an area under the receiver operating characteristic curve of 97.8% (95% CI, 86.8%–98.0%), sensitivity of 87.0% (95% CI, 67.9%–95.5%), specificity of 92.9% (95% CI, 77.4%–98.0%), precision of 0.909, recall of 0.870, and F1 score of 0.889. The heatmaps identified the critical area for prediction as the ellipsoid zone of photoreceptors and the superficial retina, which was subjected to tangential traction of the proliferative membrane. Conclusion: The novel Heterogeneous Data Fusion Net demonstrated high accuracy in the automated prediction of visual outcomes after weighing and leveraging multiple clinical parameters, including optical coherence tomography images. This approach may be helpful in establishing personalized therapeutic strategies for epiretinal membrane management.
- 1 May 2023
- Published by Wolters Kluwer Health in Retina
- Vol. 43  (5) , 767-774
PREDICTING VISUAL OUTCOME AFTER SURGERY IN PATIENTS WITH IDIOPATHIC EPIRETINAL MEMBRANE USING A NOVEL CONVOLUTIONAL NEURAL NETWORK
Purpose: To develop a deep convolutional neural network that enables the prediction of postoperative visual outcomes after epiretinal membrane surgery based on preoperative optical coherence tomography images and clinical parameters to refine surgical decision making. Methods: A total of 529 patients with idiopathic epiretinal membrane who underwent standard vitrectomy with epiretinal membrane peeling surgery by two surgeons between January 1, 2014, and June 1, 2020, were enrolled. The newly developed Heterogeneous Data Fusion Net was introduced to predict postoperative visual acuity outcomes (improvement ≥2 lines in Snellen chart) 12 months after surgery based on preoperative cross-sectional optical coherence tomography images and clinical factors, including age, sex, and preoperative visual acuity. The predictive accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve of the convolutional neural network model were evaluated. Results: The developed model demonstrated an overall accuracy for visual outcome prediction of 88.68% (95% CI, 79.0%–95.7%) with an area under the receiver operating characteristic curve of 97.8% (95% CI, 86.8%–98.0%), sensitivity of 87.0% (95% CI, 67.9%–95.5%), specificity of 92.9% (95% CI, 77.4%–98.0%), precision of 0.909, recall of 0.870, and F1 score of 0.889. The heatmaps identified the critical area for prediction as the ellipsoid zone of photoreceptors and the superficial retina, which was subjected to tangential traction of the proliferative membrane. Conclusion: The novel Heterogeneous Data Fusion Net demonstrated high accuracy in the automated prediction of visual outcomes after weighing and leveraging multiple clinical parameters, including optical coherence tomography images. This approach may be helpful in establishing personalized therapeutic strategies for epiretinal membrane management.
- Open Access
- 23 December 2022
- Published by Elsevier in Regenerative Therapy
- Vol. 22, 50-58
Anti-apoptotic and autophagic effect: Using conditioned medium from human bone marrow mesenchymal stem cells to treat human trabecular meshwork cells
- 23 December 2022
- Published by Elsevier in Regenerative Therapy
- Vol. 22, 50-58
Open AccessAnti-apoptotic and autophagic effect: Using conditioned medium from human bone marrow mesenchymal stem cells to treat human trabecular meshwork cells
- Open Access
- 9 November 2022
- Published by MDPI AG in International Journal of Molecular Sciences
- Vol. 23  (22) , 13794
Pluripotent Stem Cells in Clinical Cell Transplantation: Focusing on Induced Pluripotent Stem Cell-Derived RPE Cell Therapy in Age-Related Macular Degeneration
Human pluripotent stem cells (PSCs), including both embryonic stem cells (ESCs) and induced pluripotent stem cells (iPSCs), represent valuable cell sources to replace diseased or injured tissues in regenerative medicine. iPSCs exhibit the potential for indefinite self-renewal and differentiation into various cell types and can be reprogrammed from somatic tissue that can be easily obtained, paving the way for cell therapy, regenerative medicine, and personalized medicine. Cell therapies using various iPSC-derived cell types are now evolving rapidly for the treatment of clinical diseases, including Parkinson’s disease, hematological diseases, cardiomyopathy, osteoarthritis, and retinal diseases. Since the first interventional clinical trial with autologous iPSC-derived retinal pigment epithelial cells (RPEs) for the treatment of age-related macular degeneration (AMD) was accomplished in Japan, several preclinical trials using iPSC suspensions or monolayers have been launched, or are ongoing or completed. The evolution and generation of human leukocyte antigen (HLA)-universal iPSCs may facilitate the clinical application of iPSC-based therapies. Thus, iPSCs hold great promise in the treatment of multiple retinal diseases. The efficacy and adverse effects of iPSC-based retinal therapies should be carefully assessed in ongoing and further clinical trials.
- 9 November 2022
- Published by MDPI AG in International Journal of Molecular Sciences
- Vol. 23  (22) , 13794
Open AccessPluripotent Stem Cells in Clinical Cell Transplantation: Focusing on Induced Pluripotent Stem Cell-Derived RPE Cell Therapy in Age-Related Macular Degeneration
Human pluripotent stem cells (PSCs), including both embryonic stem cells (ESCs) and induced pluripotent stem cells (iPSCs), represent valuable cell sources to replace diseased or injured tissues in regenerative medicine. iPSCs exhibit the potential for indefinite self-renewal and differentiation into various cell types and can be reprogrammed from somatic tissue that can be easily obtained, paving the way for cell therapy, regenerative medicine, and personalized medicine. Cell therapies using various iPSC-derived cell types are now evolving rapidly for the treatment of clinical diseases, including Parkinson’s disease, hematological diseases, cardiomyopathy, osteoarthritis, and retinal diseases. Since the first interventional clinical trial with autologous iPSC-derived retinal pigment epithelial cells (RPEs) for the treatment of age-related macular degeneration (AMD) was accomplished in Japan, several preclinical trials using iPSC suspensions or monolayers have been launched, or are ongoing or completed. The evolution and generation of human leukocyte antigen (HLA)-universal iPSCs may facilitate the clinical application of iPSC-based therapies. Thus, iPSCs hold great promise in the treatment of multiple retinal diseases. The efficacy and adverse effects of iPSC-based retinal therapies should be carefully assessed in ongoing and further clinical trials.
- Open Access
- 3 June 2022
- Published by MDPI AG in Biomedicines
- Vol. 10  (6) , 1314
Widen the Applicability of a Convolutional Neural-Network-Assisted Glaucoma Detection Algorithm of Limited Training Images across Different Datasets
Automated glaucoma detection using deep learning may increase the diagnostic rate of glaucoma to prevent blindness, but generalizable models are currently unavailable despite the use of huge training datasets. This study aims to evaluate the performance of a convolutional neural network (CNN) classifier trained with a limited number of high-quality fundus images in detecting glaucoma and methods to improve its performance across different datasets. A CNN classifier was constructed using EfficientNet B3 and 944 images collected from one medical center (core model) and externally validated using three datasets. The performance of the core model was compared with (1) the integrated model constructed by using all training images from the four datasets and (2) the dataset-specific model built by fine-tuning the core model with training images from the external datasets. The diagnostic accuracy of the core model was 95.62% but dropped to ranges of 52.5–80.0% on the external datasets. Dataset-specific models exhibited superior diagnostic performance on the external datasets compared to other models, with a diagnostic accuracy of 87.50–92.5%. The findings suggest that dataset-specific tuning of the core CNN classifier effectively improves its applicability across different datasets when increasing training images fails to achieve generalization.
- 3 June 2022
- Published by MDPI AG in Biomedicines
- Vol. 10  (6) , 1314
Open AccessWiden the Applicability of a Convolutional Neural-Network-Assisted Glaucoma Detection Algorithm of Limited Training Images across Different Datasets
Automated glaucoma detection using deep learning may increase the diagnostic rate of glaucoma to prevent blindness, but generalizable models are currently unavailable despite the use of huge training datasets. This study aims to evaluate the performance of a convolutional neural network (CNN) classifier trained with a limited number of high-quality fundus images in detecting glaucoma and methods to improve its performance across different datasets. A CNN classifier was constructed using EfficientNet B3 and 944 images collected from one medical center (core model) and externally validated using three datasets. The performance of the core model was compared with (1) the integrated model constructed by using all training images from the four datasets and (2) the dataset-specific model built by fine-tuning the core model with training images from the external datasets. The diagnostic accuracy of the core model was 95.62% but dropped to ranges of 52.5–80.0% on the external datasets. Dataset-specific models exhibited superior diagnostic performance on the external datasets compared to other models, with a diagnostic accuracy of 87.50–92.5%. The findings suggest that dataset-specific tuning of the core CNN classifier effectively improves its applicability across different datasets when increasing training images fails to achieve generalization.
- 2 December 2021
- Published by Wolters Kluwer Health in Retina
- Vol. 42  (4) , 712-720
FACTORS RELATED TO UNFAVORABLE VISUAL OUTCOME AFTER IDIOPATHIC EPIRETINAL MEMBRANE SURGERY IN PATIENTS WITH GLAUCOMA
Purpose: To evaluate the visual outcomes after idiopathic epiretinal membrane surgery in glaucomatous and nonglaucomatous eyes and factors related to unfavorable outcomes in glaucomatous eyes. Methods: This was a retrospective cohort study including patients undergoing idiopathic epiretinal membrane surgery with ≥12-month follow-up. Final visual acuity at pseudophakic status was compared among groups of glaucoma, glaucoma suspect, and nonglaucoma and correlated with optical coherence tomography and visual field characteristics in patients with glaucoma. Results: Of the 314 patients enrolled, 31 had glaucoma and 22 were glaucoma suspect. Baseline visual acuity and central foveal thickness were similar across the groups. Most patients had improved/stable visual acuity postoperatively, with a lower proportion of 83.9% with glaucoma than 96.9% and 100% without glaucoma and glaucoma suspect, respectively (P = 0.002). The mean visual acuity did not change in the glaucoma group (from 6/29 to 6/23), but it improved from 6/25 to 6/12 (a gain of 16.7 approxETDRS letters) in nonglaucoma and 6/26 to 6/14 in glaucoma suspect (a gain of 14.0 approxETDRS letters) (both P < 0.001). The change of visual acuity was correlated with preoperative visual field defects (P < 0.001, r2 = 0.554). Patients with glaucoma with more advanced, fixation-threatening defects or temporally located inner nuclear layer microcysts were more likely to have worsened visual acuity. Conclusion: Visual field testing is imperative for patients with glaucoma before idiopathic epiretinal membrane surgery for outcome assessment.
- 2 December 2021
- Published by Wolters Kluwer Health in Retina
- Vol. 42  (4) , 712-720
FACTORS RELATED TO UNFAVORABLE VISUAL OUTCOME AFTER IDIOPATHIC EPIRETINAL MEMBRANE SURGERY IN PATIENTS WITH GLAUCOMA
Purpose: To evaluate the visual outcomes after idiopathic epiretinal membrane surgery in glaucomatous and nonglaucomatous eyes and factors related to unfavorable outcomes in glaucomatous eyes. Methods: This was a retrospective cohort study including patients undergoing idiopathic epiretinal membrane surgery with ≥12-month follow-up. Final visual acuity at pseudophakic status was compared among groups of glaucoma, glaucoma suspect, and nonglaucoma and correlated with optical coherence tomography and visual field characteristics in patients with glaucoma. Results: Of the 314 patients enrolled, 31 had glaucoma and 22 were glaucoma suspect. Baseline visual acuity and central foveal thickness were similar across the groups. Most patients had improved/stable visual acuity postoperatively, with a lower proportion of 83.9% with glaucoma than 96.9% and 100% without glaucoma and glaucoma suspect, respectively (P = 0.002). The mean visual acuity did not change in the glaucoma group (from 6/29 to 6/23), but it improved from 6/25 to 6/12 (a gain of 16.7 approxETDRS letters) in nonglaucoma and 6/26 to 6/14 in glaucoma suspect (a gain of 14.0 approxETDRS letters) (both P < 0.001). The change of visual acuity was correlated with preoperative visual field defects (P < 0.001, r2 = 0.554). Patients with glaucoma with more advanced, fixation-threatening defects or temporally located inner nuclear layer microcysts were more likely to have worsened visual acuity. Conclusion: Visual field testing is imperative for patients with glaucoma before idiopathic epiretinal membrane surgery for outcome assessment.
- 23 November 2020
- Published by BMJ in British Journal of Ophthalmology
- Vol. 106  (3) , 362-367
Outcomes and risk factors for failure after trabeculectomy in Taiwanese patients: medical chart reviews from 2006 to 2017
Aims To determine long-term outcomes and risk factors for failure after mitomycin C (MMC)-augmented initial trabeculectomy (IT) in Taiwanese patients. Methods We reviewed medical records of patients with glaucoma undergoing IT during December 2006–December 2016. We defined complete success as an intraocular pressure (IOP) of >5 or ≤21 mm Hg or IOP reduction of ≥20% from baseline without supplemental medications and qualified success as the aforementioned IOP levels with or without supplemental medications. Kaplan-Meier survival and Cox proportional analyses evaluated success rates and risk factors for failure, respectively. Results We enrolled 190 patients (237 eyes; mean age: 54.0±15.3 years; mean postoperative follow-up period: 68.4±35.1 months). Mean IOP and glaucoma medications decreased from 22.2±10.8 to 14.4±5.2 mm Hg (p70 years had higher complete success rates than did those aged ≤70 years. Five cases (2.11%) exhibited bleb-associated complications. Conclusion Despite satisfactory long-term success rates, most eyes needed medication for IOP control, supporting the notion of predisposed scarring vitality in patients of Chinese ethnicity following MMC-augmented trabeculectomy.
- 23 November 2020
- Published by BMJ in British Journal of Ophthalmology
- Vol. 106  (3) , 362-367
Outcomes and risk factors for failure after trabeculectomy in Taiwanese patients: medical chart reviews from 2006 to 2017
Aims To determine long-term outcomes and risk factors for failure after mitomycin C (MMC)-augmented initial trabeculectomy (IT) in Taiwanese patients. Methods We reviewed medical records of patients with glaucoma undergoing IT during December 2006–December 2016. We defined complete success as an intraocular pressure (IOP) of >5 or ≤21 mm Hg or IOP reduction of ≥20% from baseline without supplemental medications and qualified success as the aforementioned IOP levels with or without supplemental medications. Kaplan-Meier survival and Cox proportional analyses evaluated success rates and risk factors for failure, respectively. Results We enrolled 190 patients (237 eyes; mean age: 54.0±15.3 years; mean postoperative follow-up period: 68.4±35.1 months). Mean IOP and glaucoma medications decreased from 22.2±10.8 to 14.4±5.2 mm Hg (p70 years had higher complete success rates than did those aged ≤70 years. Five cases (2.11%) exhibited bleb-associated complications. Conclusion Despite satisfactory long-term success rates, most eyes needed medication for IOP control, supporting the notion of predisposed scarring vitality in patients of Chinese ethnicity following MMC-augmented trabeculectomy.
- Open Access
- 15 September 2020
- Published by Wolters Kluwer Health in Journal of the Chinese Medical Association
- Vol. 84  (1) , 108-113
XEN45 Gel Stent implantation in eyes with primary open angle glaucoma: A study from a single hospital in Taiwan
Background: To evaluate the effectiveness and safety of the XEN45 Gel Stent in East Asian patients with primary open angle glaucoma (POAG). Methods: We retrospectively reviewed 37 medically uncontrolled POAG patients who received XEN45 Gel Stent. The primary outcomes were reduction in intraocular pressure (IOP) and in the number of glaucoma medications 12 months after surgery. The secondary outcomes were requirement for intervention and further glaucoma surgery. The adverse intraoperative and postoperative events were investigated. Results: At the 12-month postoperative follow-up, the mean IOP was significantly reduced from the preoperative value of 21.7 ± 7.7 mmHg to 15.0 ± 2.0 mmHg (p = 0.001). The mean number of glaucoma medications decreased from 3.4 ± 0.9 to 1.3 ± 1.5 (p < 0.001). Seventeen patients (45.9%) required postoperative interventions. Four patients (10.8%) received additional glaucoma surgery. Postoperative IOP at month 1 was significantly associated with outcomes at the 12-month follow-up and the need for subsequent intervention and additional glaucoma surgery. Conclusion: The XEN45 Gel Stent effectively reduced the IOP values and number of glaucoma medications in East Asian patients with POAG. No major complications were observed, but almost half of the eyes in the study required intervention for wound healing modification. Postoperative IOP at month 1 was a predictor of surgical success at 12 months after surgery.
- 15 September 2020
- Published by Wolters Kluwer Health in Journal of the Chinese Medical Association
- Vol. 84  (1) , 108-113
Open AccessXEN45 Gel Stent implantation in eyes with primary open angle glaucoma: A study from a single hospital in Taiwan
Background: To evaluate the effectiveness and safety of the XEN45 Gel Stent in East Asian patients with primary open angle glaucoma (POAG). Methods: We retrospectively reviewed 37 medically uncontrolled POAG patients who received XEN45 Gel Stent. The primary outcomes were reduction in intraocular pressure (IOP) and in the number of glaucoma medications 12 months after surgery. The secondary outcomes were requirement for intervention and further glaucoma surgery. The adverse intraoperative and postoperative events were investigated. Results: At the 12-month postoperative follow-up, the mean IOP was significantly reduced from the preoperative value of 21.7 ± 7.7 mmHg to 15.0 ± 2.0 mmHg (p = 0.001). The mean number of glaucoma medications decreased from 3.4 ± 0.9 to 1.3 ± 1.5 (p < 0.001). Seventeen patients (45.9%) required postoperative interventions. Four patients (10.8%) received additional glaucoma surgery. Postoperative IOP at month 1 was significantly associated with outcomes at the 12-month follow-up and the need for subsequent intervention and additional glaucoma surgery. Conclusion: The XEN45 Gel Stent effectively reduced the IOP values and number of glaucoma medications in East Asian patients with POAG. No major complications were observed, but almost half of the eyes in the study required intervention for wound healing modification. Postoperative IOP at month 1 was a predictor of surgical success at 12 months after surgery.
- Open Access
- 17 July 2020
- Published by Wolters Kluwer Health in Journal of the Chinese Medical Association
- Vol. 83  (9) , 880-884
Prevalence of glaucoma in the elderly population in Taiwan: The Shihpai Eye Study
Background: This study aimed to evaluate the prevalence, characteristics, and the awareness status of glaucoma in an elderly Chinese population. Methods: A total of 460 individuals aged ≥72 years were enrolled in this cross-sectional community-based eye disease screening program. Glaucoma was diagnosed according to the diagnostic criteria proposed by the International Society of Geographical and Epidemiological Ophthalmology. Characteristics of subjects with primary open-angle glaucoma (POAG) and primary angle-closure glaucoma (PACG) were described and compared between groups using Mann-Whitney U and Fisher’s exact tests. Results: Forty subjects were diagnosed with glaucoma, including 17 subjects with POAG, 22 with PACG, and one with secondary glaucoma. The estimated prevalence of glaucoma, POAG, and PACG was 8.7%, 3.7%, and 4.8%, respectively. In total, 71% of the subjects with PACG and 77% of POAG subjects presented with a normal intraocular pressure (IOP) of <19 mmHg. The demographic and ocular characteristics were similar between the two groups; however, subjects with POAG had a longer axial length and smaller vertical disc diameter than those with PACG. Further, 95% of the glaucoma subjects ever visited the eye clinic, but only 32.5% of them were aware of the disease. Conclusion: Glaucoma is prevalent in the elderly population in Taiwan, but the diagnostic rate and disease awareness are low. Since most subjects with glaucoma presented with normal IOPs, optic disc evaluation is critical for diagnosing glaucoma in the elderly in Taiwan
- 17 July 2020
- Published by Wolters Kluwer Health in Journal of the Chinese Medical Association
- Vol. 83  (9) , 880-884
Open AccessPrevalence of glaucoma in the elderly population in Taiwan: The Shihpai Eye Study
Background: This study aimed to evaluate the prevalence, characteristics, and the awareness status of glaucoma in an elderly Chinese population. Methods: A total of 460 individuals aged ≥72 years were enrolled in this cross-sectional community-based eye disease screening program. Glaucoma was diagnosed according to the diagnostic criteria proposed by the International Society of Geographical and Epidemiological Ophthalmology. Characteristics of subjects with primary open-angle glaucoma (POAG) and primary angle-closure glaucoma (PACG) were described and compared between groups using Mann-Whitney U and Fisher’s exact tests. Results: Forty subjects were diagnosed with glaucoma, including 17 subjects with POAG, 22 with PACG, and one with secondary glaucoma. The estimated prevalence of glaucoma, POAG, and PACG was 8.7%, 3.7%, and 4.8%, respectively. In total, 71% of the subjects with PACG and 77% of POAG subjects presented with a normal intraocular pressure (IOP) of <19 mmHg. The demographic and ocular characteristics were similar between the two groups; however, subjects with POAG had a longer axial length and smaller vertical disc diameter than those with PACG. Further, 95% of the glaucoma subjects ever visited the eye clinic, but only 32.5% of them were aware of the disease. Conclusion: Glaucoma is prevalent in the elderly population in Taiwan, but the diagnostic rate and disease awareness are low. Since most subjects with glaucoma presented with normal IOPs, optic disc evaluation is critical for diagnosing glaucoma in the elderly in Taiwan
- Open Access
- 14 May 2020
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 15  (5) , e0233079
Deep learning assisted detection of glaucomatous optic neuropathy and potential designs for a generalizable model
To evaluate ways to improve the generalizability of a deep learning algorithm for identifying glaucomatous optic neuropathy (GON) using a limited number of fundus photographs, as well as the key features being used for classification. A total of 944 fundus images from Taipei Veterans General Hospital (TVGH) were retrospectively collected. Clinical and demographic characteristics, including structural and functional measurements of the images with GON, were recorded. Transfer learning based on VGGNet was used to construct a convolutional neural network (CNN) to identify GON. To avoid missing cases with advanced GON, an ensemble model was adopted in which a support vector machine classifier would make final classification based on cup-to-disc ratio if the CNN classifier had low-confidence score. The CNN classifier was first established using TVGH dataset, and then fine-tuned by combining the training images of TVGH and Drishti-GS datasets. Class activation map (CAM) was used to identify key features used for CNN classification. Performance of each classifier was determined through area under receiver operating characteristic curve (AUC) and compared with the ensemble model by diagnostic accuracy. In 187 TVGH test images, the accuracy, sensitivity, and specificity of the CNN classifier were 95.0%, 95.7%, and 94.2%, respectively, and the AUC was 0.992 compared to the 92.8% accuracy rate of the ensemble model. For the Drishti-GS test images, the accuracy of the CNN, the fine-tuned CNN and ensemble model was 33.3%, 80.3%, and 80.3%, respectively. The CNN classifier did not misclassify images with moderate to severe diseases. Class-discriminative regions revealed by CAM co-localized with known characteristics of GON. The ensemble model or a fine-tuned CNN classifier may be potential designs to build a generalizable deep learning model for glaucoma detection when large image databases are not available.
- 14 May 2020
- Published by Public Library of Science (PLoS) in PLOS ONE
- Vol. 15  (5) , e0233079
Open AccessDeep learning assisted detection of glaucomatous optic neuropathy and potential designs for a generalizable model
To evaluate ways to improve the generalizability of a deep learning algorithm for identifying glaucomatous optic neuropathy (GON) using a limited number of fundus photographs, as well as the key features being used for classification. A total of 944 fundus images from Taipei Veterans General Hospital (TVGH) were retrospectively collected. Clinical and demographic characteristics, including structural and functional measurements of the images with GON, were recorded. Transfer learning based on VGGNet was used to construct a convolutional neural network (CNN) to identify GON. To avoid missing cases with advanced GON, an ensemble model was adopted in which a support vector machine classifier would make final classification based on cup-to-disc ratio if the CNN classifier had low-confidence score. The CNN classifier was first established using TVGH dataset, and then fine-tuned by combining the training images of TVGH and Drishti-GS datasets. Class activation map (CAM) was used to identify key features used for CNN classification. Performance of each classifier was determined through area under receiver operating characteristic curve (AUC) and compared with the ensemble model by diagnostic accuracy. In 187 TVGH test images, the accuracy, sensitivity, and specificity of the CNN classifier were 95.0%, 95.7%, and 94.2%, respectively, and the AUC was 0.992 compared to the 92.8% accuracy rate of the ensemble model. For the Drishti-GS test images, the accuracy of the CNN, the fine-tuned CNN and ensemble model was 33.3%, 80.3%, and 80.3%, respectively. The CNN classifier did not misclassify images with moderate to severe diseases. Class-discriminative regions revealed by CAM co-localized with known characteristics of GON. The ensemble model or a fine-tuned CNN classifier may be potential designs to build a generalizable deep learning model for glaucoma detection when large image databases are not available.