Objective Assessment of Microperimetry Exam Using EEG Signals

Purpose To test the hypothesis that deep learning can decode single–trial cortical responses from electroencephalography (EEG) to individual, long–duration microperimetry (MP) stimuli, thereby providing a basis for an objective measure of visual pathway integrity. We also investigated whether occipital EEG signals could automatically register microperimetry stimuli, replacing the patient , s manual button press to improve exam reliability. Methods We collected pilot EEG data from two healthy participants (12 trials total) during microperimetry (Nidek MP1) at low (3.2 cd/m 2 ) and high (127 cd/m 2 ) stimulus intensities. Occipital EEG signals (8–channel OpenBCI) were bandpass filtered (4–49 Hz), z–score normalized, and segmented into 600 ms epochs time–locked to stimuli. A bidirectional LSTM (Bi–LSTM) deep neural network then classified each epoch as stimulus or non-stimulus. Results In seven low-intensity trials (16 dB), the model achieved up to ~77.4% accuracy in distinguishing stimulus vs. non–stimulus EEG segments using the O1/O2 channels. In five high–intensity trials (0 dB), accuracy reached ~80%. Occipital electrodes (O1, O2) outperformed parietal or combined parieto–occipital montages for detecting the flash stimuli. Consistent with the neuroanatomy of the primary visual cortex, classification performance was maximal using the bilateral occipital electrodes (O1, O2). Conclusion Integrating occipital EEG with microperimetry provides a proof–of–concept objective framework that reduces reliance on subjective patient feedback and could enhance diagnostic precision. While our pilot study was limited by a small sample and lack of precise EEG–stimulus synchronization, the findings warrant further refinement for clinical translation. Translational Relevance Enabling automated stimulus registration can improve microperimetry , s reliability in pediatric and challenging populations.