A self‐gated 4D‐MRI sequence for internal target volume delineation in liver: Phantom and pre‐clinical validation

Background The poor soft tissue resolution of four‐dimensional computed tomography (4D‐CT) limits its utility in delineating liver cancer target volumes. Purpose To compare the consistency between four‐dimensional magnetic resonance imaging (4D‐MRI) using T1‐weighted (T1w) radial stack‐of‐stars (SOS) gradient echo (GRE) sequences and 4D‐CT in assessing tumor motion and morphology, for defining internal target volume in liver tumor radiotherapy. Materials and Methods Position and geometric accuracy and the impact of baseline drift between 4D‐MRI (using T1w radial SOS GRE sequence) and 4D‐CT were evaluated using a motion phantom. Ten primary liver cancer patients were included for comparison. Volume, centroid position, dice similarity coefficient (DSC), and 95% Hausdorff distance (HD95) were assessed. Results Baseline drift did not significantly affect target motion or morphology descriptions in 4D‐MRI (normal vs. drift: p = 0.139, DSC = 0.94 ± 0.01, HD95 = 1.82 ± 0.43) or 4D‐CT ( p = 0.051, DSC = 0.95 ± 0.03, HD95 = 1.59 ± 0.71), but significantly impacted target volume delineation (4D‐CT: p = 0.040; 4D‐MRI: p = 0.024). Phantom studies showed consistent motion amplitude measurements between modalities for small motion patterns (2 s/10 mm; p = 0.192), while 4D‐MRI underestimated amplitudes in large patterns (4 s/20 mm; p = 0.012), consistent with patient data (small: p = 0.453; large: p = 0.004). Volume analysis revealed 4D‐MRI underestimated phantom volumes by 6.33% (2 s/10 mm) and 6.10% (4 s/20 mm) versus ground truth, contrasting with 4D‐CT's overestimations (2.01% and 2.51%). Clinically, 4D‐CT‐derived gross tumor volumes (GTVs)/internal tumor volumes (ITVs) exceeded 4D‐MRI measurements, particularly in large motion patterns (GTVs: (11.79 ± 13.13)%; ITVs: (22.08 ± 18.59)%) versus small patterns (GTVs: (9.70 ± 9.10)%; ITVs: (11.01 ± 8.05)%). Temporal stability metrics were comparable between modalities for small motion (phantom DSC p = 0.770, HD95 p = 0.131; patient DSC p = 0.156), with overall GTV/ITV spatial agreement (DSC: 0.79 ± 0.06 vs. 0.83 ± 0.04 for 4D‐CT vs. 4D‐MRI). Conclusion The results show 4D‐MRI with a T1w radial SOS GRE sequence matches 4D‐CT's accuracy in target motion characterization and morphological delineation, while providing precise tumor delineation for primary liver cancer patients.