Weakly Supervised Segmentation and Classification of Alpha-Synuclein Aggregates in Brightfield Midbrain Images

TL;DR

The paper addresses automated analysis of alpha-synuclein aggregates in brightfield midbrain WSIs by proposing a weakly supervised segmentation pipeline that fuses stain-separation and attention-based methods, followed by a self-supervised, nearest-neighbors retrieval approach to build a six-class morphology classifier. The segmentation relies on Vahadane-based stain decomposition and ViT-based attention refined by CRF, while classification uses a ResNet-50 (and other backbones) trained on exemplars derived from unlabeled data. The method achieves strong segmentation performance and up to 80% balanced accuracy in morphologic classification, enabling scalable morphometric and spatial studies of aggregates and their interactions with glial cells. The approach offers a practical path toward large-scale characterization of alpha-synuclein pathology in histology, with potential extensions to more classes and unlabeled data utilization.

Abstract

Parkinson's disease (PD) is a neurodegenerative disorder associated with the accumulation of misfolded alpha-synuclein aggregates, forming Lewy bodies and neuritic shape used for pathology diagnostics. Automatic analysis of immunohistochemistry histopathological images with Deep Learning provides a promising tool for better understanding the spatial organization of these aggregates. In this study, we develop an automated image processing pipeline to segment and classify these aggregates in whole-slide images (WSIs) of midbrain tissue from PD and incidental Lewy Body Disease (iLBD) cases based on weakly supervised segmentation, robust to immunohistochemical labelling variability, with a ResNet50 classifier. Our approach allows to differentiate between major aggregate morphologies, including Lewy bodies and neurites with a balanced accuracy of $80\%$. This framework paves the way for large-scale characterization of the spatial distribution and heterogeneity of alpha-synuclein aggregates in brightfield immunohistochemical tissue, and for investigating their poorly understood relationships with surrounding cells such as microglia and astrocytes.

Figures (2)

• Figure 1: Overview of the segmentation pipeline for alpha-synuclein aggregates in brightfield immunohistochemistry.(A) Input tile undergoes semantic segmentation using a vision transformer classifier. The resulting attention map is refined with a fully connected Conditional Random Field (CRF) and combined with stain decomposition to isolate magenta-stained regions corresponding to alpha-synuclein. (B) Instance segmentation is performed on the refined binary mask to extract individual aggregates, followed by post-processing steps including label association and small object filtering. Output: instance segmentation mask $\textbf{S}$ of alpha-synuclein aggregates. (C) Final output: $256 \times 256$ image crops centered on each segmented alpha-synuclein aggregate, used for downstream classification, with segmentation contours in green.
• Figure 2: Image retrieval procedure. (A) Retrieval of images from the unlabeled dataset $\Omega$ that are most similar to the query input image $q_i$, using a $k$-nearest neighbors search on embeddings generated by the neural network $f_{\theta}$. (B) Representative examples of alpha-synuclein aggregates for each class, extracted using the described method.