Semi-automated, machine-learning based segmentation and analysis of mouse brain stroked tissues
Loading...
Date
2023-07-05
Authors
Γέμου, Ιωάννα
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
An unbiased, automated, fast, and reliable method for analyzing brain lesions after an ischemic stroke is currently lacking, with existing approaches being time-consuming, prone to human error, or overly simplistic. In this study, we developed an innovative tool that leverages deep-learning techniques to enable semi-automated stroke analysis in brain slices stained with the Nissl protocol. Our approach incorporates the use of the comprehensive Allen Mouse Brain Atlas, providing a standardized spatial framework for precise analysis. Accessible through a user-friendly web interface, our tool allows researchers and medical professionals to easily upload their brain slice images and obtain automated stroke analysis results. We employed an off-the-shelf image segmentation model and created a custom made dataset specifically for stroke lesion detection, in order to fine tune the model, enhancing the accuracy of our analysis. Additionally, image registration techniques were utilized to extract vital anatomical information. By aligning stained brain slices with a reference atlas, we ensured consistent spatial mapping of pathological regions, facilitating comprehensive analysis within a standardized anatomical context. Our tool has been extensively validated using a large dataset, demonstrating high accuracy and reliability. The semi-automated analysis accurately identifies and measures pathological regions, overcoming the constraints associated with manual analysis techniques. This innovative methodology represents a substantial progress in the examination of brain lesions following an ischemic stroke, offering an impartial and automated solution for researchers and medical professionals working with stained brain sections.
Description
Keywords
Machine learning, Deep learning, Mouse brain, Ischemic stroke, Stroke lesion segmentation, Stroked region identificication