Multiclass semantic segmentation mediated neuropathological readout in Parkinson's disease

Abstract

Automated segmentation of anatomical sub-regions with high precision has become a necessity to enable the quantification and characterization of cells/ tissues in histology images. An automated model to do this task is currently unavailable. One area of the brain which requires precise sub-region segmentation and downstream analysis is Substantia Nigra (SN). The loss of dopaminergic (DA) neurons in SN is the primary endpoint for majority of Parkinson's disease (PD) preclinical studies. The scientists rely on manually segmenting anatomical sub-regions of the brain which is extremely time-consuming and prone to labeler-dependent bias. In this study, we employed a UNet-based architecture to segment two sub-regions of SN-dorsal tier of substantia nigra pars compacta (SNCD) and reticulata (SNr). We compared model performance with various combinations of encoders, image sizes and sample selection techniques. The model is trained on approximately one thousand annotated 2D brain images stained with Nissl/ Haematoxylin and Tyrosine Hydroxylase enzyme (TH, indicator of dopaminergic neuron viability). The framework's output are: segmentation of SNr and SNCD irrespective of the tissue staining, quantitative readout for TH intensity indicating DA health status in the segmented regions. With the availability of training data, this model can be expanded to other 2D sub-region segmentation tasks. The shorter turnaround time, high accuracy and unbiased data output of this model will fulfill the ever increasing demands of data analysis in PD preclinical research.

Keywords

• Digital pathology
• Parkinson's disease
• Machine learning
• Automated segmentation
• Convolutional neural network