In the Journals

Artificial intelligence-based algorithm screens for diabetic retinopathy

A novel artificial intelligence-based algorithm proved effective as a tool to automate screening for diabetic retinopathy, according to a study.

Deep learning methods were used to process 75,137 color fundus images of patients with diabetes obtained from the EyePACS public data set. The artificial intelligence model was trained and tested to detect and differentiate the features of healthy eye fundi from eye fundi with DR.

A tree-based classification model labeled the eyes with no DR as 0 and the eyes with DR of any severity as 1; the eyes labeled as 1 were referred for further care. The performance of the algorithm was evaluated against two other public databases: Messidor-2, containing 1,748 fundus images from four French institutions, and e-ophtha of the French Research Agency, containing 463 images.

“Our model achieved a 0.97 [area under the receiver operating characteristic curve] with a 94% and 98% sensitivity and specificity, respectively. ... Testing against the independent Messidor-2 and e-ophtha databases achieved a 0.94 and 0.95 AUC score, respectively,” the study authors said.

“The results attained indicate the high efficacy of our computer-aided model in providing efficient, low-cost and objective DR diagnostics without depending on clinicians to examine and grade images manually,” they wrote.

The algorithm, they noted, requires no costly, sophisticated technology, and it can be run on common personal computers and smartphones. Implementation on a global basis could help reduce the rate of diabetes-related blindness, they said. by Michela Cimberle

Disclosure: Gargeya reports patent application No. 62383333, with a filing date of Sept. 2, 2016.

A novel artificial intelligence-based algorithm proved effective as a tool to automate screening for diabetic retinopathy, according to a study.

Deep learning methods were used to process 75,137 color fundus images of patients with diabetes obtained from the EyePACS public data set. The artificial intelligence model was trained and tested to detect and differentiate the features of healthy eye fundi from eye fundi with DR.

A tree-based classification model labeled the eyes with no DR as 0 and the eyes with DR of any severity as 1; the eyes labeled as 1 were referred for further care. The performance of the algorithm was evaluated against two other public databases: Messidor-2, containing 1,748 fundus images from four French institutions, and e-ophtha of the French Research Agency, containing 463 images.

“Our model achieved a 0.97 [area under the receiver operating characteristic curve] with a 94% and 98% sensitivity and specificity, respectively. ... Testing against the independent Messidor-2 and e-ophtha databases achieved a 0.94 and 0.95 AUC score, respectively,” the study authors said.

“The results attained indicate the high efficacy of our computer-aided model in providing efficient, low-cost and objective DR diagnostics without depending on clinicians to examine and grade images manually,” they wrote.

The algorithm, they noted, requires no costly, sophisticated technology, and it can be run on common personal computers and smartphones. Implementation on a global basis could help reduce the rate of diabetes-related blindness, they said. by Michela Cimberle

Disclosure: Gargeya reports patent application No. 62383333, with a filing date of Sept. 2, 2016.