Though many advancements in the fields of imaging and genetic testing have occurred in the preceding decades, their remains difficulty in predicting accurately diagnosing rare eye diseases. The electroretinogram is the only objective functional measure of retinal function and is typically reliant on the use of subjective peaks for stratification. However voltage peaks are unable to capture waveform shape, and it’s possible these structures many enable improved diagnostic stratification. Specifically discrete wavelet transform (DWT) is a non-redundant time-frequency method capable of capture waveform shape using selective features. We leverage DWT in pattern ERG (PERG) uniquely fitted to capture macular cone and retinal ganglion cells in patients with inherited retinal diseases (IRDs), and optic neuropathies (ONs). We identify a set of waveform indices using a repeat monte-carlo simulation that routinely capture canonical peaks in the native time-domain recording and exhibit improved functional stratification between healthy controls and IRDs. We then validate these same measures in a ON cohort while also demonstrating efficacy in retinal ganglion cell indices and improved stratification. Lastly, we show top performing predictive models in all scenarios utilize DWT as a part of their feature set. By using time-frequency specific features that can capture waveform shape and structure, we demonstrate a method and set of measures that improve better explain differences between healthy controls, and pathologies like IRDs and ON.
Yousif Shwetar (Fri,) studied this question.