Meeting News

Test combining fundus, OCT imaging improves AMD diagnosis

SAN FRANCISCO — A screening strategy combining fundus and OCT imaging improved the accuracy of age-related macular degeneration classification, according to data presented at the American Academy of Ophthalmology annual meeting.

The bimodal deep convolutional neural networks (DCNN) “combo” test could be used to help facilitate screening in community medical centers and better differentiate wet AMD from polypoidal choroidal vasculopathy (PCV), Zhiyan Xu, of the Chinese Academy of Medical Sciences and the department of ophthalmology at Peking Union Medical College Hospital in Beijing, China, reported.

“As we all know, PCV and wet AMD have similar manifestations on the color fundus graph,” she said. “Recent studies have claimed that PCV belongs to one specific subtype of neovascular AMD. The treatment strategies for PCV and wet AMD are different. It is really necessary for us to find the accurate diagnosis.”

Xu and colleagues compared the performance of the bimodal DCNN combo test with a unimodal DCNN test using OCT alone (DCNN-OCT test), a unimodal DCNN test using fundus alone (DCNN-fundus test) and human expert opinion. The researchers assessed the tests using fundus and OCT images that were retrospectively collected from 827 participants whose diagnoses were previously confirmed by the gold standard of diagnosis.

The bimodal DCNN-combo test “presented the most promising performance,” Xu said. It demonstrated 87.4% accuracy, 88.8% sensitivity and 95.6% specificity, and was the only machine test to outperform human experts, which comparatively demonstrated 86% accuracy, 88.1% sensitivity and 95.2% specificity. The bimodal DCNN-combo test also achieved a “near-perfect agreement” with the gold standard of diagnosis (K score = 0.841), according to the researchers.

When comparing only the unimodal tests, Xu and colleagues found that DCNN-OCT was more effective than DCNN-fundus imaging in terms of accuracy (83.2% vs. 75%), sensitivity (86% vs. 75%) and specificity (94.1% vs. 91.7%).

Despite these promising findings, Xu acknowledged that the data from this single-center study are limited.

“Thus, multiple-center studies involving more clinical data and various disease types are warranted,” she said. – by Stephanie Viguers

Reference: Xu Z, et al. Automated classification of polypoidal choroidal vasculopathy and other subtypes of AMD using bimodal deep convolutional neural networks. Presented at: American Academy of Ophthalmology annual meeting; Oct. 11-15, 2019; San Francisco.

Disclosure: Xu reports receiving grant support from the Chinese Academy of Medical Sciences (CAMS) Initiative for Innovative Medicine and National Natural Science Foundation of China.

SAN FRANCISCO — A screening strategy combining fundus and OCT imaging improved the accuracy of age-related macular degeneration classification, according to data presented at the American Academy of Ophthalmology annual meeting.

The bimodal deep convolutional neural networks (DCNN) “combo” test could be used to help facilitate screening in community medical centers and better differentiate wet AMD from polypoidal choroidal vasculopathy (PCV), Zhiyan Xu, of the Chinese Academy of Medical Sciences and the department of ophthalmology at Peking Union Medical College Hospital in Beijing, China, reported.

“As we all know, PCV and wet AMD have similar manifestations on the color fundus graph,” she said. “Recent studies have claimed that PCV belongs to one specific subtype of neovascular AMD. The treatment strategies for PCV and wet AMD are different. It is really necessary for us to find the accurate diagnosis.”

Xu and colleagues compared the performance of the bimodal DCNN combo test with a unimodal DCNN test using OCT alone (DCNN-OCT test), a unimodal DCNN test using fundus alone (DCNN-fundus test) and human expert opinion. The researchers assessed the tests using fundus and OCT images that were retrospectively collected from 827 participants whose diagnoses were previously confirmed by the gold standard of diagnosis.

The bimodal DCNN-combo test “presented the most promising performance,” Xu said. It demonstrated 87.4% accuracy, 88.8% sensitivity and 95.6% specificity, and was the only machine test to outperform human experts, which comparatively demonstrated 86% accuracy, 88.1% sensitivity and 95.2% specificity. The bimodal DCNN-combo test also achieved a “near-perfect agreement” with the gold standard of diagnosis (K score = 0.841), according to the researchers.

When comparing only the unimodal tests, Xu and colleagues found that DCNN-OCT was more effective than DCNN-fundus imaging in terms of accuracy (83.2% vs. 75%), sensitivity (86% vs. 75%) and specificity (94.1% vs. 91.7%).

Despite these promising findings, Xu acknowledged that the data from this single-center study are limited.

“Thus, multiple-center studies involving more clinical data and various disease types are warranted,” she said. – by Stephanie Viguers

Reference: Xu Z, et al. Automated classification of polypoidal choroidal vasculopathy and other subtypes of AMD using bimodal deep convolutional neural networks. Presented at: American Academy of Ophthalmology annual meeting; Oct. 11-15, 2019; San Francisco.

Disclosure: Xu reports receiving grant support from the Chinese Academy of Medical Sciences (CAMS) Initiative for Innovative Medicine and National Natural Science Foundation of China.

    See more from Discoveries from AAO: AMD