Career Center AES Connect

Abstracts

Bridging the Gap: AI-Enhanced Seizure-Like Event Detection and Drug Response Evaluation via Microelectrode Arrays

Abstract number : 2.3
Submission category : 7. Anti-seizure Medications / 7E. Other
Year : 2023
Submission ID : 697
Source : www.aesnet.org
Presentation date : 12/3/2023 12:00:00 AM
Published date :

Authors :
Presenting Author: Duong Nhu, PhD – Monash University

Ana Atonic-Baker, PhD – Research Fello, Department of Neuroscience, Monash University; Zonguyan Ge, PhD – Monash University; Patrick Kwan, PhD – Professor, Department of Neuroscience, Monash University; Hannah Leeson, PhD – University of Queensland; Selin Pars, PhD – PhD, University of Queensland; Lata Vadlamudi, PhD – University of Queensland; Ernst Wolvegang, PhD – University of Queensland; Muhammad Shahid, PhD – Research Fellow, Department of Neuroscience, Monash University

Rationale: Seizure-like events (SLEs) observed in vitro models have emerged as valuable biomarkers for assessing drug response in epileptic patients (Dulla et al., 2018; Grainger et al., 2018). However, the high signal complexity of these events poses a challenge in their identification. Our study proposes a deep-learning model to automatically detect SLEs from microelectrode array (MEA) recordings obtained from 2D-neurons induced Pluripotent Stem Cells (iPSC) of epileptic patients.

Methods: The objective of this study is to distinguish SLEs from normal neuron firing events. We used 4-Aminopyridine (4-AP) to induce seizures in neurons generated from an epileptic patient. Spike bursts of this recording were considered SLEs. We also recorded signals from healthy cell lines and used the spike bursts from these normal neuron firing events. In our dataset, there were 15,182 SLEs and 16,913 normal firing events. We partitioned the dataset into three distinct sets: a training set comprising 60% of the data, a validation set with 10%, and a test set holding 30%. We extracted a diverse range of dynamical and time-series features. These included fractal dimensions, entropies such as singular-value decomposition, sample, permutation, and spectral entropies, as well as statistical features of amplitudes containing kurtosis, skewness, mean, standard deviation, and variance. Additionally, we employed Catch22 analysis, spike duration, spike frequency, and burst duration. The deep-learning model utilised in this study was a five layer multilayer perceptron classifier. The inputs were normalised to between zero and one by using min-max normalization.

To assess the applicability of our approach in evaluating drug response, we first recorded baseline activity (without 4-AP and drugs) from 2D neurons using MEA, and then conducted additional recordings treated with varying dosages of Carbamazepine (CBZ) (100 μM, 33 μM, and 10 μM). Furthermore, to investigate the effectiveness of seizure prevention, we introduced 4-Aminopuridine (4-AP) in combination with each dosage of CBZ. This experimental setup allowed us to examine the impact of different drug concentrations on the occurrence and mitigation of SLEs.


Results: We achieved an AUC of 0.99 (95% CI: 9.98-0.99) on the test set. We observed that the number of detected SLEs dropped after adding CBZ where 100 μM showed the highest reduction. After adding 4-AP, the number of SLEs increased. However, compared to the seizure recording (4-AP-only), 100 μM had the highest difference in the number of SLEs. We might conclude that 100 μM of CBZ had the best effect on the epileptic patient in our study. The results are summarized in Table 1 and Table 2.

Conclusions: Our study proposed an effective deep-learning model to automatically detect SLEs from MEA recordings. The preliminary results indicate the feasibility of using this model in assessing antiseizure drug response by analyzing the changes in the number of detected SLEs. We will extend the iPSC pipeline to more patients and anti-seizure drugs and show the results at the time of the presentation.

Funding:

Medical Research Future Fund



Anti-seizure Medications