Report on the quantitative intra-voxel incoherent motion diffusion MRI reconstruction grand challenge

Abstract

Purpose

The 2024 quantitative intra-voxel incoherent motion diffusion MRI (IVIM-dMRI) reconstruction grand challenge aimed to benchmark and advance reconstruction algorithms for extracting quantitative tissue parameters from diffusion MRI (dMRI) data. Focusing on the IVIM model, the challenge aimed to improve the accuracy and robustness of clinical parameter estimation, addressing key barriers to broader clinical adoption.

Methods

Participants were tasked with reconstructing fractional perfusion, pseudo-diffusion coefficient, and true diffusion coefficient from simulated $k$-space data based on realistic digital VICTRE phantoms. The challenge consisted of three phases: training, validation, and testing, with a focus on evaluating reconstruction performance using relative root mean square error (rRMSE). Both traditional optimization and deep learning (DL)-based methods were allowed.

Results

The challenge attracted 42 teams from six countries, with seven progressing to the final phase. The rRMSE ranged in [0.0345, 1.24]. The top-performing algorithm employed a cascaded U-Net architecture for image denoising and parameter fitting. Overall, the competition highlighted the potential of advanced methodologies, particularly DL, in addressing complex inverse problems in medical imaging.

Conclusion

The IVIM-dMRI grand challenge demonstrated significant advancements in the accuracy and robustness of dMRI reconstruction. Although the simulation-based approach provided a controlled environment, future efforts must address real-world complexities to ensure clinical applicability.