Computer Aided Diagnosis of Multiple Eye Fundus Diseases From Color Fundus Photograph

As a prospective clinical trial, This study enjoys multicentric, blind film reading, self-control and superiority test design. In total, 1,500 retinal fundus images from 750 individuals in need of fundus examination (one image for every single eye) were selected. Then a test group, along with a control group was set up in our study. For the test group, ophthalmologists read images with the aid of the assistant software(MCS). In contrast, the same work in the control group was finished by ophthalmologists independently. Meanwhile, the gold standard were obtained from the cooperation of senior ophthalmologists. Diagnoses of both groups were compared with those of the gold standard, thus the investigators could evaluated the safety and effectiveness of this assistant software in diagnosis.

The primary endpoint of this study is the superiority of the consistency rate of the test group. A diagnosis for an image is consistent if it gives the same negative result as the reference standard, or reveals any one condition indicated by the reference standard. The consistency rate is the rate of consistent diagnoses for all the involved images. One control group is designed, where each doctor reads and diagnoses, and give at most 3 possible conditions for each image. In the test group, doctors do the same thing with the help of this software. The investigators in the test group and control group are the same and they are chosen from ophthalmologists with 1~3 years experience. The reference standard of each fundus image is collaboratively given by retinal specialists/fellows from 5 centers. The investigator of XieHe center is the arbitrator if full consensus cannot be reached for any image during the building of reference standard.

Inclusion criteria

• Age between 18 and 75.
• Anyone need to take fundus photograph in clinical.
• Understand the study and volunteer to sign the informed consent.
• For fundus images of participants, the optic disc, fovea, the upper and lower vessel bow should be included in the fundus field.

Exclusion criteria

• Participants has any eye that cannot take fundus photos.
• Participants have joined or is participating in other clinical trial within one month.
• Participants who have any other issue that cannot be enrolled.
• Participants with cloudy refractive media that cannot take fundus photos or get clouding fundus photos.
• Participants with low quality fundus photos like incompetent vision field, overexposed/underexposed, out of focus, too many shadow or dirties and so on.