A weakly supervised model for incomplete multimodal MRI synthesis with tumor-aware approach

Area of Science:

• Medical imaging
• Artificial intelligence
• Neuro-oncology

Background:

• Multimodal magnetic resonance images (MRIs) offer valuable insights into brain tumors.
• Missing MRI modalities in clinical practice hinder comprehensive tumor analysis.
• Accurate tumor information is crucial for effective clinical diagnosis.

Purpose of the Study:

• To develop a high-precision algorithm for synthesizing missing MRI modalities.
• To generate accurate, tumor-specific information from incomplete multimodal MRI data.
• To enhance data availability for brain tumor diagnosis.

Main Methods:

• Proposed a novel weakly supervised MRI synthesis model: TAM-DAM-GAN.
• Integrated tumor-aware and detail adjustment mechanisms for enhanced tumor generation.
• Utilized weak labels and pixel-level attention maps to guide and refine synthesis.

Main Results:

• Evaluated synthesis quality across four tasks (e.g., FLAIR-to-T1, T1-to-T2).
• TAM-DAM-GAN demonstrated superior qualitative and quantitative performance on the BRATS2015 dataset.
• Improved PSNR in tumor regions and boosted tumor segmentation accuracy by 10% when using synthesized data.

Conclusions:

• The developed algorithm enhances cross-modality synthesis accuracy for incomplete multimodal MRI, particularly in tumor regions.
• Provides more dependable and comprehensive data for clinical diagnosis and scientific research.
• Facilitates improved understanding and treatment of brain tumors.