Blood vessel analysis beyond segmentation: development of flexible approaches for characterizing vascularization morphology

• Grant number: 23/03975-4
• Support Opportunities: Scholarships in Brazil - Doctorate
• Start date: September 01, 2023
• End date: December 31, 2026
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Cesar Henrique Comin
• Grantee: Matheus Viana da Silva
• Host Institution: Centro de Ciências Exatas e de Tecnologia (CCET). Universidade Federal de São Carlos (UFSCAR). São Carlos , SP, Brazil

Abstract

Blood vessel characterization is of fundamental importance for studying pathologies that influence hemodynamics, and is also crucial for understanding the development and activity of the brain. Recent advances in microscopy improved the quality and resolution of the imaged samples, and the automated morphometry of blood vessels became crucial in this current scenario, given the increasingly larger data volumes. Usually, the main focus of novel blood vessel characterization methods lies in the segmentation step. However, recent works indicate the possibility of having morphometry biases even in situations where a segmentation of high accuracy is present. In this project, we propose to build a blood vessel morphometry pipeline that incorporates the error of the morphological characteristics of interest in the optimization of the segmentation step, ensuring high reliability to automated analyses of blood vessels. We plan to do this by optimizing segmentation neural networks using loss functions that weight the segmentation error by the influence that each pixel of the label has on the calculation of the metrics of interest. To support our experiments, we will label and organize an extensive dataset comprising blood vessels imaged by confocal microscopy. Subsequently, we will explore the usage of data augmentation methodologies based on morphology, which aims to expand the heterogeneity of our training datasets by means of modifying the morphological characteristics of the images. We expect that the developed methodologies will substantially increase the accuracy of studies that rely on high-precision automated blood vessel morphometry, such as in the neurosciences. Further, we hope that the public availability of our dataset will aid novel researches that aim to characterize blood vessels with machine learning. (AU)