Accelerating breast MRI acquisition with generative AI models

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Breast MRI screening is increasingly recommended, especially for women with dense breasts.
• Accelerated image acquisition is crucial for improving MRI accessibility and patient experience.
• Current breast MRI reconstruction methods face challenges in balancing speed and image quality.

Purpose of the Study:

• To evaluate the efficacy of score-based diffusion models for accelerating breast MRI reconstruction.
• To assess the image quality and diagnostic value of reconstructions generated by these models.
• To investigate the performance of the models at various undersampling factors.

Main Methods:

• A score-based diffusion model was trained on 9,549 breast MRI examinations.
• The model was used to reconstruct undersampled MRI images with acceleration factors of 2, 5, and 20.
• Two radiologists assessed the overall quality and diagnostic value of reconstructed images on an independent test set of 100 examinations.

Main Results:

• The score-based model successfully reconstructed T1- and T2-weighted breast MRI images with high fidelity.
• At an acceleration factor of 2, images were rated nearly indistinguishable from originals by 100% (radiologist 1) and 99% (radiologist 2) of cases.
• Performance decreased with higher acceleration factors, achieving 88% and 70% for factor 5, and 5% and 21% for factor 20, respectively.

Conclusions:

• Score-based diffusion models demonstrate potential for high-fidelity breast MRI reconstruction, even at moderate acceleration factors.
• Further research with larger datasets is necessary to confirm diagnostic quality at higher acceleration levels.
• This approach could significantly enhance the efficiency and accessibility of breast MRI examinations.