Medical Image Imputation from Image Collections.

Area of Science:

• Medical Imaging
• Computational Anatomy
• Artificial Intelligence

Background:

• Clinical brain MRI scans often have large inter-slice spacing due to acquisition time constraints.
• Sparse scans limit the capture of anatomical details, hindering computational analysis.
• Existing super-resolution methods struggle with generalizing to diverse clinical image datasets.

Purpose of the Study:

• To develop a generative model for enhancing resolution in undersampled brain MRI scans.
• To enable the application of existing image analysis algorithms to sparse clinical MRI data.
• To improve the quality of clinical brain MRI scans for better anatomical analysis.

Main Methods:

• Introduced a generative model to capture fine-scale anatomical structures across subjects.
• Developed an algorithm to fill missing data in MRI scans with large inter-slice spacing.
• Utilized a generative adversarial network approach for image synthesis.

Main Results:

• The proposed method significantly outperforms state-of-the-art upsampling super-resolution techniques.
• Generated high-resolution, anatomically plausible images consistent with clinical brain MRI scans.
• Demonstrated improved performance in facilitating subsequent image analysis.

Conclusions:

• The developed algorithm effectively addresses the challenge of sparse MRI data.
• This approach enhances the utility of clinical MRI scans for research and diagnostics.
• The method promises to unlock new possibilities for computational analysis of undersampled neuroimaging data.