Microscopy Image Dataset for Deep Learning-Based Quantitative Assessment of Pulmonary Vascular Changes.

Area of Science:

• Cardiovascular Pathology
• Medical Imaging Analysis
• Computational Biology

Background:

• Pulmonary hypertension (PH) involves structural changes in pulmonary vessels, impacting prognosis.
• Collagen deposition thickens vessel walls, worsening PH and reducing treatment efficacy.
• Accurate histological analysis of pulmonary vessels is crucial but time-consuming.

Purpose of the Study:

• To present a novel dataset for deep learning-based pathological assessment of pulmonary circulation vessels.
• To facilitate the development of automated software for analyzing histological microphotographs.
• To address the limitations of manual quantitative measurements in PH research.

Main Methods:

• Creation of a dataset comprising 609 microphotographs of pulmonary vessels.
• Inclusion of expert-derived quantitative measurements and corresponding annotated images.
• Demonstration of a deep learning pipeline utilizing U-Net for semantic segmentation of vascular regions.

Main Results:

• A comprehensive dataset of annotated pulmonary vessel microphotographs is now available.
• A U-Net based deep learning model was successfully applied for vascular region extraction.
• The dataset supports the development of automated histological analysis tools for PH.

Conclusions:

• The presented dataset and deep learning approach can significantly streamline the analysis of pulmonary vessel histology.
• Automated analysis holds potential for improving diagnostic accuracy and efficiency in PH research and clinical practice.
• This resource will accelerate the development of novel computational tools for cardiovascular pathology assessment.