Related Experiment Video
Updated: Nov 22, 2025
10:25
Brain Infarct Segmentation and Registration on MRI or CT for Lesion-symptom Mapping
Published on: September 25, 2019
48.9K
RatLesNetv2: A Fully Convolutional Network for Rodent Brain Lesion Segmentation
Juan Miguel Valverde1, Artem Shatillo2, Riccardo De Feo1,3,4
1A.I. Virtanen Institute for Molecular Sciences, University of Eastern Finland, Kuopio, Finland.
Frontiers in Neuroscience
|January 8, 2021
Summary
RatLesNetv2, a novel deep learning tool, accurately segments rodent brain lesions from MRI scans. This automated method improves segmentation quality and reproducibility in preclinical research.
Area of Science:
- Neuroimaging
- Artificial Intelligence
- Biomedical Engineering
Background:
- Accurate segmentation of brain lesions in rodents is crucial for preclinical studies, particularly in drug development for focal cerebral ischemia.
- Manual segmentation is time-consuming, subjective, and prone to variability, impacting research reproducibility.
Purpose of the Study:
- To introduce RatLesNetv2, a fully convolutional neural network (ConvNet) for automated segmentation of rodent brain lesions in magnetic resonance (MR) images.
- To evaluate the performance of RatLesNetv2 against existing methods on a large dataset and assess its potential for improving lesion segmentation in preclinical research.
Main Methods:
- Development of RatLesNetv2, a residual-block-enhanced autoencoder-like ConvNet architecture designed for end-to-end training on 3D MR images without preprocessing.
- Evaluation on a dataset of 916 T2-weighted rat brain MRI scans from 671 rats across nine lesion stages, relevant to focal cerebral ischemia studies.
- Comparative analysis against three other medical image segmentation ConvNets, assessing Dice coefficient, segmentation realism, compactness, and Hausdorff distance.
Main Results:
- RatLesNetv2 achieved Dice coefficient values comparable to or exceeding those of other evaluated ConvNets.
- Segmentations generated by RatLesNetv2 were more realistic, compact, and exhibited fewer holes and lower Hausdorff distances compared to other methods.
- The performance of RatLesNetv2 surpassed the inter-rater agreement observed for manual segmentations.
Conclusions:
- RatLesNetv2 offers a robust and automated solution for rodent brain lesion segmentation, significantly reducing manual workload.
- The tool enhances the reproducibility and accuracy of lesion quantification in preclinical research, particularly for stroke and drug development studies.
- RatLesNetv2 is publicly available, promoting wider adoption and advancement in neuroimaging analysis.
