Abstracts

Rationale:
In recent years, multielectrode arrays (MEAs) have emerged as a powerful tool for recording electrical activity from human brain slices, providing valuable insights into the underlying mechanisms of epilepsy. The objective of our study is to identify the optimal parameters, address artifacts, and enhance signal analysis through post-processing techniques.

Methods:
We analyzed 16-minute long 60-channel MEA recordings (25 kHz sampling rate) recorded at the Department of Physiology at the Faculty of Medicine of Masaryk University in cooperation with Brno Epilepsy Center (St Anne’s Hospital) in 2022. Measurements were performed on resected hippocampal tissue from 3 patients with drug-resistant focal epilepsy. An automatic spike detector was employed to detect neuronal activity. To optimize the detection, the amplitude threshold was systematically varied across values of 5, 5.5, 6, and 6.5. The threshold corresponds to a multiplication of a median power of a signal. Furthermore, different high-pass filter settings of 100, 200, and 300 Hz were tested. False detections were characterized by the occurrence of detections at the same time or in a range of ±3 consecutive samples in at least 25 % of channels. The active electrodes were identified within a given time interval using a 0.1 Hz spike rate threshold. We determined the total count of active electrodes and visualized the spike rate for each active electrode. Comparisons of active electrode quantities across amplitude thresholds were made, and spike sorting and visual inspection of detected spikes were then used for final parameter determination.

Results:
Due to limited tissue viability, recordings from three patients out of 17 were included for analysis, focusing on patient number 16. The best parameters were then determined as 6.5 amplitude threshold and a 300 Hz cut-off frequency. In the example recording, the number of active electrodes at a cut-off frequency of 300 Hz and removed false detections was 31 at a threshold of 5, 27 at 5.5, 23 at 6, and 22 at 6.5, detailed in Figure 1. When using a cut-off frequency of 100 Hz with the presence of false detection, the number of active electrodes for the respective thresholds was 58, 55, 41, and 24. In an example with non-optimal parameters (Figure 1B), the boxplot graph displays an extremely skewed spike rate distribution characterized by a low median and a significant number of outliers. In addition, the total count of 58 active electrodes  (97%) is highly uncommon. These observations suggest that the chosen parameters are not suitable for further analysis. Figure 2 highlights a visible difference in neural activity across channels between A and B. The removal of false detections in B results in a cleaner representation of the neural activity.

Conclusions:
We have determined that the desired combination of parameters for the analysis includes a 6.5 amplitude threshold and a 300 Hz cut-off frequency for the high-pass filter with the proposed post-processing steps. The proposed approach should reduce false detections and increase the probability of detecting true neuronal activity.

Funding:
Project n. NU22-08-00278 of the Ministry of Health of the Czech Republic.