Calibration-Free Neurochemical Sensing: CNN and ANN Modeling of FSCV Signals for Dopamine, Serotonin, and Toxic Metals

Neurotransmitter dysregulation and exposure to toxic metals such as copper and cadmium are increasingly recognized as key contributors to neurodegenerative diseases. Fast-scan cyclic voltammetry (FSCV) with carbon fiber microelectrodes (CFMs) enables real-time monitoring of neurochemicals, particularly dopamine (DA) and serotonin (5-HT). However, FSCV data are complex due to overlapping electrochemical responses, making quantitative analysis challenging. Further, calibration typically relies on in vitro measurements, which fail to replicate the biological complexity of in vivo environments, often leading to inaccuracies in concentration estimates and limiting the translational impact of FSCV. To address these challenges, we developed a machine learning framework integrating convolutional neural networks (CNNs) and artificial neural networks (ANNs) for advanced analysis of FSCV data. CNNs differentiated the voltammetric signals of DA, 5-HT, Cu²⁺, and Cd²⁺ in tris buffer, extracting analyte-specific features that are difficult to resolve by conventional means. These features were then used to train ANNs for concentration prediction, offering a calibration-free alternative to traditional approaches. CNN models achieved high classification accuracy in distinguishing overlapping signals, while ANN models provided low mean absolute errors in concentration prediction. Together, this AI-driven strategy reduces dependence on external calibrations, enhances analyte discrimination, and improves the reliability of FSCV for in vivo applications. By enabling accurate monitoring of neurotransmitters and toxic metals, this work provides new opportunities to study the interplay between neurotransmission and metal toxicity in neurodegeneration.