ML-driven segmentation of histologic microvascular features

Customer: Kuzbass Cardiology Center (Kemerovo, Russia)

Summary: This project focuses on developing advanced machine learning (ML) tools to perform histological analysis of tissue-engineered vascular grafts (TEVGs). Histopathological examination is crucial for assessing the outcomes of implanted medical devices, particularly for understanding the complex architecture of regenerated tissues. To address this need, we compiled a dataset of 104 whole slide images (WSIs) from TEVGs implanted in sheep carotid arteries for six months. The dataset was processed and annotated to identify nine key histological features. Using automated slicing and manual annotation, we segmented 1401 patches from the WSIs. Six deep learning models (U-Net, LinkNet, FPN, PSPNet, DeepLabV3, and MA-Net) were rigorously tuned and evaluated for their performance in segmenting and quantifying these features. All models achieved mean Dice Similarity Coefficients (DSC) above 0.823, with MA-Net achieving the highest mean DSC of 0.875. An ensemble of MA-Net, DeepLabV3, and FPN further improved the average DSC to 0.889. This study demonstrates the potential of ML-driven segmentation in histological analysis of TEVGs, providing a robust tool for detecting essential histological features and advancing tissue engineering research.

Collaborators: Kirill Klyshnikov (Kuzbass Cardiology Center, Kemerovo, Russia), Evgeny Ovcharenko (Kuzbass Cardiology Center, Kemerovo, Russia), Anton Kutikhin (Kuzbass Cardiology Center, Kemerovo, Russia)

Project type: Research

Media:  Journal paper,  GitHub repo

Figure 3. Comparison of models for microvascular segmentation in tissue-engineered vascular grafts.