Automated Abnormality Detection in Patient Retinal Function: A Deep Learning-Powered Electroretinogram Analysis System

The electroretinogram (ERG) is an ophthalmic electrophysiology test designed to objectively measure the electrical response of the photoreceptor cells in the human retina. The analysis of the ERG is highly useful in evaluating various retinal diseases arising from both inherited and acquired disorders. However, analyzing ERG test results can be a time-intensive process for electrophysiologists, requiring meticulous examination of recorded ERG waveforms for accurate diagnosis and assessment. In this paper, we present a novel deep learning-based ERG analysis system to automatically detect abnormalities in the retinal function of patients. To train and validate the proposed system, a dataset of 5,640 ERG tracings recorded from 470 patients, along with diagnoses from experienced electrophysiologists, was built. Our results indicate that the proposed system demonstrates effectiveness in detecting anomalies in retinal function on full-field ERGs with an F1-score of 77.59%, and an inference time of 1.2 milliseconds. This system can be used as an effective tool to assist electrophysiologists and ophthalmologists by providing rapid and reliable automated assessments.