Deep Learning-based System and AIDS-related Cytomegalovirus Retinitis

Kuifang Du

Status: Completed

Conditions: Cytomegalovirus Retinitis

Study type: Observational

Funder types: Other

Identifiers

NCT04831333

20210331001

Details and patient eligibility

About

Ophthalmological screening for cytomegalovirus retinitis (CMVR) for HIV/AIDS patients is important. However, the manual screening with fundus imaging is laborious and subjective.

Deep learning (DL) system has been developed for the automated detection of various eye diseases with high accuracy and efficiency, including diabetic retinopathy, glaucoma, age-related macular degeneration (AMD), papilledema, lattice degeneration and retinal breaks, from ocular fundus photographs. UWF imaging is a relatively new imaging modality for DL system but has also shown extraordinary talents in automatic retinal analysis With the press for routine CMVR screening in AIDS patients and the great capacity of DL system, the use of deep learning (DL) system to AIDS-related CMVR with Ultra-Widefield (UWF) fundus images is promising.

The investigators previously developed a DL system to detect AIDS-related CMVR. For further evaluating the applicability of the DL system, a prospective dataset is needed.

Enrollment: 50 patients

Sex: All

Ages: 18+ years old

Volunteers: No Healthy Volunteers

Inclusion criteria

The UWF images from HIV/AIDS patients.

Exclusion criteria

1. The UWF images would be excluded if all three human graders gave different diagnosis.
2. The UWF images with poor quality would be excluded.

Trial design

50 participants in 3 patient groups

Active CMVR

Description:

The UWF images of cytomegalovirus retinitis (CMVR) included various patterns: hemorrhagic necrotizing lesion, granular lesion, frosted branch angiitis, and optic neuropathy lesion. Active CMVR lesion was defined as obvious opacity (mild, moderate, severe, very severe)

Inactive CMVR

Description:

Inactive CMVR lesion was defined as a lack of opacity or questionable/equivocal activity.

Non-CMVR

Description:

The non-CMVR images included normal retina and other retinopathies such as HIV-related microvascular retinopathy, diabetic retinopathy, retinal detachment, vitreous hemorrhage.