A Fuzzy-Based Approach for Interpretable Spike Detection in Living Neural Biocomputers

The convergence of neuroscience and artificial intelligence has led to the development of “living biocomputers,” where biological neural networks are cultured on Micro-Electrode Array (MEA) chips. This creates hybrid biocomputing platforms that blur the boundaries between biology and technology. A critical challenge in this domain is the accurate and interpretable detection of neural spikes from MEA recordings. While black-box models like deep neural networks have achieved good accuracy, their lack of transparency hinders scientific inquiry. Revealing spatial-temporal patterns across electrode arrays can inform electrode placement strategies, stimulation protocols, and our understanding of how biological neural networks process information. This approach could bridge the gap between understanding the organization of neurons, their activity and the computing task in hand and ultimately allow living neural biocomputers to have a solution to the current “black-box” biocomputing approaches. We adopt a fuzzy logic framework because it can handle the biological uncertainty and signal variability inherent in MEA recordings, enabling more faithful spike discrimination than rigid threshold-based or crisp machine learning approaches. This paper presents a fuzzy rule-based classifier that combines high performance with human-readable interpretability for spike detection in neural computing. Our approach integrates ANNIGMA (Artificial Neural Network for Input Gain and Measurement Analysis)-based feature selection to identify the most salient spike characteristics and a genetic algorithm to optimize the fuzzy rule base. We validated our model using multi-chip pooled validation protocol, achieving a peak F1 score on testing data of 97.74% on a two-chip experiment and 96.12% on a six-chip experiment. The resulting fuzzy rules are linguistically interpretable, allowing researchers to understand the underlying logic of spike detection. This work can be a step forward towards building trustworthy neural biocomputers with direct applications to drug discovery, disease modeling, and understanding the fundamentals of neural computation.