Accuracy and Feasibility of AI-Assisted OCT in Retinal Disease

AI Generated Summary*

Screening in four Shanghai communities tested a deep-learning tool embedded in an OCT device that flags 15 retinal abnormalities on 12 by 9 mm posterior pole scans. Of 954 eyes from 477 adults, 878 cleared quality control. Residents, attending doctors and retinal specialists graded cases, the specialists serving as reference. No detachment cases arose; across the other 14 conditions AUCs ran from 0.891 for drusen to 0.997 for cystoid macular edema, with sensitivity 87.65 to 100 percent and specificity 80.12 to 99.41 percent. Agreement with specialists was closer for the algorithm than for the junior or senior groups, mean kappa 0.731 versus 0.579 and 0.707. Scans averaged 18.4 seconds per eye. Limitations include one device type and one district.

*This summary was generated by AI and is published unedited. Oku does not alter these summaries. It may contain errors or omissions and is provided for general informational purposes only. Oku does not guarantee its accuracy, completeness, or reliability. For authoritative information, please refer to the original, peer-reviewed article.