The Classification of Flash Visual Evoked Potential Based on Deep Learning

Abstract

Background: Visual electrophysiology is an objective visual function examination widely used in clinical work, which can objectively evaluate the function of the retina or optic nerve, and objectively reflect the corresponding changes of the disease according to the waveform change.In the visual electrophysiological examination, FVEP varies greatly among individuals, resulted in the different waveforms in different normal subjects. Moreover, most of the FVEP waves labeling were performed by the professional clinical technicians. Those labeling may has bias due to individual variation in subjects, incomplete clinical examination data, differences professional skills, personal habits and other factors. The labeling results can be very different, then disturb the clinical diagnosis.In the past, computer algorithms were mainly based on manually set rules for diagnosis, which was limited by the complex and volatile clinical conditions. This time, through the retrospective study of big data, artificial intelligence algorithm is used to maintain high generalization ability in complex situations and improve the accuracy of pre-screening. Methods: A novel multi-input neural network based on convolution and confidence branching (MCAC-Net) for RP recognition and out-of-distribution detection is proposed. The MCAC-Net with global and local feature extraction is designed for the FVEP signal that has different information locally and globally, and a confidence branch is added for out-of-distribution sample detection. For the proposed manual features, a new input layer is added. Results: The model is verified by the clinically collected FVEP dataset, and an accuracy of 90.7% is achieved in the classification task and 93.3% in the out-of-distribution detection task. Conclusion: We built a deep learning-based FVEP classification algorithm that promises to be an excellent tool for screen of RP diseases by using FVEP signals.