GLAUCOMA FIELD DEFECT CLASSIFIER (GFDC)

 

Glaucoma Field Defect Classifier (GFDC) is a web application that automates glaucoma grading based on Hodapp-Parish-Anderson (HPA) criteria without requiring patient-identifiable data.

In a study presented by Cambridge University, UK, researchers, the GFDC web-application output matched the ground truth defined by human researchers applying HPA criteria for every perimetry result.

Perimetry appraisal using GFDC was significantly faster than a manual application of HPA criteria. However, the duration of manual appraisal exhibited greater variation, with some manual assessments being faster than any GFDC-facilitated assessment.

Simple thresholds for mean deviation and central global plot decibel values are explicitly coded into the algorithm.

To interpret pattern deviation plots, a computer vision algorithm is designed to identify plot boundaries and result points. A matrix is generated based on the pattern deviation identified at each result point, which is then used to apply encoded HPA criteria.

This web-application has the potential to facilitate the incorporation of HPA-based visual field assessment at scale.

100% sensitivity for the detection of any glaucomatous field defect and a severe field defect suggests that no patients would be dismissed as a false negative by GFDC, maximizing safety with deployment.

100% specificity for detecting severe field defects indicates that the algorithm can identify patients at high risk or with significant deficits without compromising efficiency by including other patients as false positives.

Adopting standard criteria for visual field appraisal in glaucoma clinics would ameliorate one of the most severe sources of arbitrary discrepancy in diagnosis, assessment of progression, prognosis, and identification of vision impairment.

Using an explicitly coded computer vision algorithm reduces the time required for clinicians to leverage validated criteria, overcomes black box limitations associated with machine learning approaches, and minimizes the possibility of erroneous decisions made for uninterpretable reasons.

SOFTWARE AVAILABLE THROUGH:

https://github.com/RohanSanghera/gfdc

REFERENCE:

Thirunavukarasu AJ, Jain N, Sanghera R, Lattuada F, Mahmood S, Economou A, Yu HCY, Bourne R. A validated web-application (GFDC) for automatic classification of glaucomatous visual field defects using Hodapp-Parrish-Anderson criteria. NPJ Digit Med. 2024 May 18;7(1):131. doi: 10.1038/s41746-024-01122-8. PMID: 38762669; PMCID: PMC11102533.