TrueTH: A user-friendly deep learning approach for robust dopaminergic neuron detection

Abstract

Parkinson’s disease (PD) entails the progressive loss of dopaminergic (DA) neurons in the substantia nigra pars compacta (SNc), leading to movement-related impairments. Accurate assessment of DA neuron health is vital for research applications. Manual analysis, however, is laborious and subjective. To address this, we introduce TrueTH, a user-friendly and robust pipeline for unbiased quantification of DA neurons. Existing deep learning tools for tyrosine hydroxylase-positive (TH+) neuron counting often lack accessibility or require advanced programming skills. TrueTH bridges this gap by offering an open-sourced and user-friendly solution for PD research. We demonstrate TrueTH’s performance across various PD rodent models, showcasing its accuracy and ease of use. TrueTH exhibits remarkable resilience to staining variations and extreme conditions, accurately identifying TH+ neurons even in lightly stained images and distinguishing brain section fragments from neurons. Furthermore, the evaluation of our pipeline’s performance in segmenting fluorescence images shows strong correlation with ground truth and outperforms existing models in accuracy. In summary, TrueTH offers a user-friendly interface and is pretrained with a diverse range of images, providing a practical solution for DA neuron quantification in Parkinson’s disease research.