Two-step machine learning method for the rapid analysis of microvascular flow in intravital video microscopy

Area of Science:

• Physiology
• Biomedical Engineering
• Medical Imaging

Background:

• Microvascular blood flow is vital for tissue health and function.
• Diseases often impair microcirculation, necessitating detailed study.
• Current analysis of Intravital Video Microscopy (IVM) data is manual, time-consuming, and prone to errors.

Purpose of the Study:

• To develop an automated, accurate method for analyzing microvascular blood flow using IVM images.
• To overcome the limitations of manual image analysis in microcirculation research.
• To apply machine learning for functional assessment of in vivo microvascular dynamics.

Main Methods:

• A two-step image processing algorithm was developed.
• The first step employed a modified vessel segmentation algorithm to identify blood vessels.
• The second step utilized a 3D Convolutional Neural Network (CNN) to determine blood flow within identified vessels.

Main Results:

• The algorithm successfully automated the functional analysis of IVM images.
• The two-step approach achieved high accuracy (83%) in assessing microvascular blood flow.
• This represents the first use of machine learning for in vivo functional microvascular analysis.

Conclusions:

• The developed AI algorithm offers an efficient and accurate alternative to manual analysis of microvascular blood flow.
• This technology has the potential to accelerate research in microcirculation and related diseases.
• Automated analysis of IVM data using CNNs can significantly improve the study of in vivo microvascular function.