ESTRO 2024 - Abstract Book

S3925

Physics - Image acquisition and processing

ESTRO 2024

Figure 2 shows example images from a patient with a T4 tonsil tumour and adjacent node. Correction with topup LS and topup Jac partially compensated for the opposing distortion seen in the two EPI acquisitions, but SPLICE appears most similar to the tumour shape seen on T2W. Central slice median (IQR) ADC tumour values were: 0.68 (0.40), 0.87 (0.30), 0.74 (0.26), 0.75 (0.26), 0.69 (0.38) µm 2 /ms for EPI, SPLICE, topup LS and topup Jac and topup b0 , respectively. As with the phantom data, SPLICE gave higher ADC values than EPI. In a second patient with a T3 base of tongue tumour, all topup methods failed to correct for significant signal pile-up in the acquired EPI images, which may be due to movement between reversed PE acquisitions; signal pile-up was not observed with SPLICE. Correction of EPI images with topup LS and topup Jac reduced the variability in the distortions that were observed in the three orthogonal diffusion gradient directions ( G ); SPLICE was inherently less affected by these G -specific distortions.

Conclusion:

This work provides the first comparison of two approaches to obtaining geometrically robust H&N DW images on an MR-Linac, and contributes to the technical validation of DW-MRI for adaptive radiotherapy. Our experience suggests SPLICE is more robust than distortion-corrected EPI in terms of geometric accuracy, but overestimates ADC. These considerations may be important for trials which adapt radiotherapy based on ADC values. Ongoing work is evaluating the repeatability of the two approaches and quantifying geometric accuracy.

Keywords: Diffusion MRI, distortion correction, H&N cancer

References:

1. McDonald et al., Radiother Oncol 2023;185:109717. 2. Andersson et al., Neuroimage Clin 2003;20:870–888. 3. Winter et al., Acta Oncologica 2017;56:1659-1663. 4. Schakel et al., Med Phys 2017;44:4188-4193. 5. https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/topup; accessed October 2023. 6. Jenkinson et al., NeuroImage 2012;62:782-790.

Acknowledgements