Evaluating a Low-Cost Replacement for the Humphrey Visual Field Test Using AI-Based Image Analysis of Binocular Visual Field (BVF) Data
Loading...
Date
2025-05
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
The Ohio State University
Abstract
Visual field testing is essential for diagnosing and managing conditions like glaucoma, yet the current clinical standard, the Humphrey Visual Field (HVF) test, is costly, time-intensive, and not always easily accessible. This project focuses on validating a low-cost, paper-based alternative developed by the Ooi Lab known, a unique Binocular Visual Field (BVF) test, where patients manually draw regions of vision loss. To analyze and compare these subjective BVF results with HVF outcomes, I developed an AI-assisted image processing pipeline and graphical user interface (GUI) to quantify spatial agreement. The objectives of this study were to (1) build a GUI-based tool to extract BVF contours and overlay them with HVF pattern deviation maps, (2) compute agreement metrics such as Intersection over Union (IoU) and Hausdorff distance across 16 patient eye samples, and (3) integrate a lightweight decision tree classifier to assess whether spatial drawing errors could explain observed BVF-HVF mismatches. Using classical computer vision methods, I accurately segmented hand-drawn BVF regions and aligned them with clinical HVF images using field-of-view scaling and manual adjustment controls. The model was trained on image-derived spatial features and revealed that centroid shift (a measure of how far off a BVF drawing was from the HVF result) was the strongest predictor of misalignment. This confirmed that many mismatches stemmed from patient spatial reproduction error rather than perceptual inaccuracies. The analysis revealed moderate-to-strong spatial agreement across most cases, supporting the idea that BVF drawings can meaningfully reflect visual field deficits when interpreted using structured analysis. This work demonstrates that patient-drawn BVF results, when combined with AI-based analysis tools, can offer meaningful diagnostic value. The methods developed here lay the groundwork for future touchscreen-based BVF tests with real-time feedback and support scalable, accessible vision screening for patients unable to complete traditional HVF testing.
Description
Keywords
Binocular Visual Field Test, Humphrey Visual Field Test, Optometry, Computer Vision, Machine Learning