Deep learning framework for generating nonfat-suppressed MR images from fat-suppressed scans

This technology is a deep learning model that uses a U-Net neural network to generate nonfat-suppressed breast MRI images from standard fat-suppressed scans, allowing for diagnostic evaluation.

Unmet Need: Lack of paired fat and nonfat images in breast MRI

T1-weighted fat-suppressed imaging is the current standard for detecting lesions and guiding diagnosis for breast cancer patients. Non-fat-suppressed images are often not acquired in scanning protocols, even though they can be essential in identifying fat necrosis, fatty lesions, and background parenchymal enhancement, all of which are key biomarkers of breast cancer risk. The lack of both fat- and nonfat-suppressed images in standard MRI protocols limits treatment, necessitating new computational methods to improve diagnostic accuracy and efficiency without increasing scan time or patient burden.

The Technology: Model converting fat-suppressed images to nonfat-suppressed breast MRIs

This technology offers a modular deep learning framework called Sat2Nu for generating nonfat-suppressed images from fat-suppressed MRI scans. It employs a U-Net architecture optimized to process both image types, allowing for accurate training even with limited datasets. Validation studies demonstrated low reconstruction error in the generated images compared to true nonfat-suppressed images, supporting its potential for improving MRI workflows without requiring additional scan acquisitions, particularly in breast imaging applications.

Applications:

  • Breast cancer imaging and diagnostics
  • Deep learning models for magnetic resonance imaging (MRI)
  • Background parenchymal enhancement (BPE) quantification
  • Research model for studying breast cancer
  • Improve abbreviated MRI workflows

Advantages:

  • Generates nonfat-suppressed images without additional scans
  • Enables training with limited datasets
  • Reduces scan times and improves MRI workflow efficiency
  • Compatible with existing MRI systems and imaging protocols

Lead Inventor:

Despina Kontos, Ph.D.

Related Publications:

Tech Ventures Reference: