Extended Modality Propagation: Image Synthesis of Pathological Cases

Area of Science:

• Medical image analysis
• Computational imaging
• Artificial intelligence in medicine

Background:

• Synthesizing multi-modal medical images, especially for pathological cases, is challenging.
• Existing generative models often focus on healthy anatomy or lack pathological specificity.
• The need for large, diverse datasets for training and validating medical image algorithms is critical.

Purpose of the Study:

• To develop a novel generative model for synthesizing multi-modal medical images of pathological cases from a single label map.
• To extend the Modality Propagation strategy for pathological image synthesis, termed Extended Modality Propagation.
• To estimate the uncertainty associated with the synthesized medical images.

Main Methods:

• The model integrates a generative model for label fusion and segmentation with an iterative strategy for spatially-coherent scan synthesis.
• The Extended Modality Propagation method is applied to pathological case synthesis.
• Image synthesis uncertainty is quantified.

Main Results:

• The model successfully synthesizes multi-modal Magnetic Resonance Imaging (MRI) of glioma-bearing brains.
• Validation was performed on a Multimodal Brain Tumor Image Segmentation challenge dataset.
• Performance was compared against state-of-the-art glioma image synthesis techniques.

Conclusions:

• The developed generative model enables the creation of large synthetic datasets of pathological medical images.
• This approach can significantly benefit the training, validation, and benchmarking of medical image processing algorithms.
• The Extended Modality Propagation method offers a promising direction for pathological image synthesis.