Quantification of dopaminergic neurons in the substantia nigra pars compacta (SNc) of animal models is important for understanding the pathogenesis of Parkinson’s disease (PD). However, conventional manual cell counting method requires the time and effort, and has limited reproducibility due to inter- and intra-examiner variability. Here, we demonstrate that a commercially available convolutional neural network–based artificial intelligence (AI) counting method (TruAI, OLYMPUS, Tokyo, Japan) enables robust and reproducible quantification of TH-positive dopaminergic neurons in mouse, marmoset, and human SNc samples when compared with conventional manual counting. AI-based counting showed a strong correlation with manual counting across mouse, marmoset, and human samples. Good agreement between AI-based and manual counting was observed in mouse and marmoset samples, supporting the applicability of this approach for cross-species quantification of dopaminergic neurons. In the mouse model treated with α-syn preformed fibrils (PFFs), AI-based counting detected a significant reduction in TH-positive neurons consistent with expert manual counting. Non-experts exhibited greater intra-examiner variability than an expert, indicating that the reliability of manual counting depends on experience. Overall, AI-based quantification provides a robust and objective approach for TH-positive cell counting and may improve reproducibility in dopaminergic neuron analysis, particularly for non-expert users and cross-species studies of PD.
Kunugitani et al. (Tue,) studied this question.