Modality propagation: coherent synthesis of subject-specific scans with data-driven regularization.

Area of Science:

• Medical Imaging
• Computational Biology
• Machine Learning

Background:

• Current medical image synthesis methods are often application-specific.
• Label propagation (LP) is successful for segmentation but not for continuous data synthesis.
• Generalizing LP for diverse modality synthesis is needed.

Purpose of the Study:

• To develop a general database-driven framework for coherent synthesis of subject-specific scans across desired modalities.
• To generalize the patch-based label propagation (LP) strategy for continuous data synthesis.
• To apply the generalized LP framework for abnormality detection in multi-channel MRI.

Main Methods:

• A generalized patch-based label propagation (LP) framework for modality synthesis.
• A data-driven regularization scheme with iterative search-and-synthesis.
• Spatio-population search space restriction for efficient population data leverage.
• Synthesis of T2 and Diffusion Tensor Imaging - Fractional Anisotropy (DTI-FA) from T1-weighted Magnetic Resonance Imaging (MRI) input.

Main Results:

• Demonstrated high-quality synthesis of T2 and DTI-FA MRI signals from T1 input.
• Successfully applied the modality synthesis framework for abnormality detection in multi-channel brain tumor MRI.
• The generalized LP framework enables synthesis beyond categorical segmentation labels.

Conclusions:

• The proposed framework offers a generalized approach to medical image synthesis using label propagation.
• This method facilitates coherent synthesis of various MRI modalities.
• The framework shows promise for applications like abnormality detection in clinical settings.