Active Learning Harnessed to Help Measure Contrast Sensitivity Function in nnAMD

Contrast sensitivity function (CSF) measured using a novel active learning method was significantly decreased in early non-neovascular age-related macular degeneration (nnAMD), according to the results of a study presented by Neal S. Patel, MD, of Massachusetts Eye and Ear, at the American Academy of Ophthalmology 2020 Virtual meeting.

According to Dr. Patel, assessments of contrast sensitivity better correlate with functional vision used in day-to-day activities than visual acuity. Indeed, CSF is perhaps an even more accurate measurement of visual function but is difficult to quantify.

The absence of a reliable and clinically applicable test has hindered the adoption of CSF testing in routine clinical practice, Dr. Patel and colleagues wrote.

Therefore, they developed the Quick Contrast Sensitivity Function (qCSF) method to test CSF. This method uses artificial intelligence with active learning to make the measurement process quicker and more accurate. The model draws from 128 possible contrasts and 19 possible spatial frequencies with a total bank of more than 2,400 candidate test items.

According to Dr. Patel, it has high sensitivity, good test-test-reliability, takes about five minutes per eye, and maximizes information gain.

To test this new method, Dr. Patel and colleagues measured CSF using qCSF in patients with nnAMD (129 eyes) compared with healthy controls (133 eyes); 31 eyes had early nnAMD, 88 had intermediate nnAMD, and nine had advanced nnAMD. The primary outcome was area under the log contrast sensitivity function curve (AULCSF) and contrast sensitivity threshold at spatial frequencies ranging from one to 18 cycles per degree.

Multivariate regression analysis showed that in early nnAMD CSF thresholds at low spatial frequencies were significantly decreased despite no difference in best corrected visual acuity (P<0.01 for all). For intermediate and advanced nnAMD, CSF thresholds at low spatial frequencies and AULCSF were decreased compared with controls (P<0.05 for all).

There were no significant differences in higher spatial frequencies identified. In addition, AULCSF was able to differentiate between nnAMD stages.

Based on this, “CSF may emerge as a promising visual functional endpoint in clinical practice and future nnAMD clinical trials,” Dr. Patel and colleagues said on their poster. “Future investigations may include longitudinal studies with qCSF tested over progression of macular disease.”