Prob-BBDM: A Probabilistic Brownian Bridge Diffusion Model for MRI sequence image-to-image translation.

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Image Synthesis

Background:

• AI-driven image synthesis is advancing medical imaging applications.
• Acquiring multiple MRI sequences is resource-intensive and time-consuming.
• There is a need for efficient methods to synthesize MRI sequences.

Purpose of the Study:

• To propose a novel image-to-image translation model for synthesizing MRI sequences from 2D axial slices.
• To leverage Brownian Bridge Diffusion Models (BBDM) with a variational encoder for enhanced synthesis quality.
• To evaluate the performance, efficiency, and generalizability of the proposed model.

Main Methods:

• Developed a Probabilistic-BBDM (Prob-BBDM) integrating a variational encoder-guided diffusion mechanism.
• Trained and evaluated the model on the BraTS 2021 dataset for MRI sequence synthesis.
• Assessed clinical utility using synthesized slices for tumor segmentation with a pre-trained model.

Main Results:

• Prob-BBDM achieved superior performance with up to 88.46% SSIM and 26.09 dB PSNR.
• The synthesis process required only 4 diffusion steps, demonstrating computational efficiency.
• Synthesized slices used for tumor segmentation achieved a Dice score of 88.71% and HD95 of 3.49mm.
• Consistent performance was observed on an external third-party dataset, confirming generalizability.

Conclusions:

• Prob-BBDM offers high-quality, efficient, and generalizable MRI synthesis.
• The model preserves critical diagnostic information, showing clinical utility.
• This work presents a promising advancement in AI-driven medical image translation.