Spectroscopic and deep learning-based approaches to identify and quantify cerebral microhemorrhages
Christian Crouzet1,2, Gwangjin Jeong3, Rachel H Chae4
1Beckman Laser Institute and Medical Clinic, University of California-Irvine, Irvine, CA, USA.
Scientific Reports
|May 22, 2021
Summary
Digital pathology methods can automate the identification of cerebral microhemorrhages (CMHs) in brain tissue. Deep learning and ratiometric analysis offer precise and accurate quantification, improving data analysis speed.
Area of Science:
- Neuroscience
- Pathology
- Biomedical Imaging
Background:
- Cerebral microhemorrhages (CMHs) are linked to cerebrovascular diseases and cognitive decline.
- Current manual analysis of Prussian blue-stained histological sections is subjective and time-consuming.
- Automated quantification methods are needed to improve efficiency and reduce variability.
Purpose of the Study:
- To develop and compare digital pathology approaches for identifying and quantifying CMHs.
- To assess the accuracy and precision of ratiometric analysis, phasor analysis, and deep learning methods.
- To evaluate the potential of these methods for accelerating CMH data analysis.
Main Methods:
- Three digital pathology techniques were applied: ratiometric RGB analysis, phasor analysis, and deep learning (mask R-CNN).
- Methods were tested on Prussian blue-stained brain sections from a preclinical mouse model of inflammation-induced CMHs.
- Ground truth was established through independent manual annotations by four users.
Main Results:
- Deep learning and ratiometric analysis outperformed phasor analysis in CMH identification compared to ground truth.
- The deep learning approach demonstrated the highest precision.
- Ratiometric analysis offered versatility with maintained accuracy, though with slightly less precision.
Conclusions:
- Digital pathology approaches significantly enhance the speed of CMH image analysis.
- Deep learning and ratiometric analysis provide accurate and precise quantification of CMHs.
- These automated methods can overcome the limitations of manual analysis, improving research efficiency.
