Deep learning-based quantification of brain atrophy using 2D T1-weighted MRI for Alzheimer's disease classification.
Chae Jung Park1, Yu Hyun Park2,3,4, Kichang Kwak5
1Research Institute, National Cancer Center, Goyang, Republic of Korea.
Frontiers in Aging Neuroscience
|August 29, 2024
Summary
Deep learning models using accessible 2D T1 MRI scans accurately detect brain atrophy in dementia of the Alzheimer's type (DAT), matching the precision of 3D scans for cost-effective diagnosis.
Area of Science:
- Neuroimaging
- Artificial Intelligence in Medicine
- Neurology
Background:
- Brain atrophy assessment is vital for diagnosing neurodegenerative diseases.
- Current 3D T1 MRI methods are time-consuming and costly.
- Developing accessible alternatives for brain atrophy quantification is essential.
Purpose of the Study:
- To introduce deep learning algorithms for brain atrophy quantification using 2D T1 MRI.
- To achieve cost-effective differentiation of dementia of the Alzheimer's type (DAT) from cognitively unimpaired (CU).
- To maintain or exceed the performance of 3D T1 imaging and predict AD-specific atrophy changes.
Main Methods:
- Trained deep learning models on cerebrospinal fluid (CSF) volumes from 924 participants (478 CU, 446 DAT) using 2D T1 images.
- Compared 2D T1-derived metrics with those from 3D T1 images.
- Assessed diagnostic performance using receiver operating characteristic analysis and Pearson's correlation.
Main Results:
- Strong correlations (r=0.805-0.971) were found between 2D and 3D T1-derived CSF volumes.
- 2D T1 algorithms accurately differentiated DAT from CU (AUC=0.873), comparable to 3D T1.
- High correlations were observed for AD-specific atrophy similarity, W-scores, and Brain Age Index (BAI) using 2D T1.
Conclusions:
- Deep learning analysis of 2D T1 MRI is a feasible and accurate method for brain atrophy assessment.
- This approach offers comparable diagnostic precision to 3D T1 imaging.
- 2D T1 MRI provides a reduced-cost, time-efficient alternative for dementia diagnosis.
