Preprocessing and decoding neural traces of serial biases in working memory from intracranial electrophysiological recordings in humans

Abstract

Working memory is a fundamental cognitive process enabling short-term maintenance and manipulation of information. In recent years, behavioral studies have revealed biases affecting working memory contents. It has been observed that information maintained in our working memory buffer (e.g., spatial location, color, auditory frequency) shifts towards previously observed positions, an effect known as serial dependency bias, and its magnitude correlates with conditions as schizophrenia and NMDAR encephalitis. This project sought to characterize the neural correlates of working memory and serial dependency biases using intracranial electroencephalography (iEEG), an invasive technique with high temporal and spatial resolution. Nevertheless, iEEG analyses present several challenges associated with the signal’s nature that need addressing. In this TFG, multiple artifact detection and cleaning methods were developed and evaluated on two criteria: how many trials and electrodes it preserved, and how accurately it detected current trial information and serial biases from preceding trials. The custom automated approach provided marginally higher decoding accuracy at the cost of greater data loss, whereas the manual cleaning delivered equivalent performance while preserving more data, rendering it preferable for wholebrain analyses. After artifact removal, multiple analyses confirmed that information regarding both current stimuli and past trials could be decoded from distributed iEEG patterns. Although limited by a small epilepsy cohort, these findings offer practical guidance for future iEEG studies by demonstrating the trade-offs between data preservation and signal clarity and suggesting that serial dependency signals may be distributed across brain regions. This project represents the first investigation of working memory’s mechanisms using human iEEG.

Date of Award	2025
Original language	English
Awarding Institution	Universitat de Barcelona (UB)
Supervisor	Alexis Perez Bellido (Director)