KEY POINTS
- Investigators benchmarked 10 commercially available or adapted head phantoms ranging from a homogeneous sphere to anthropomorphic skull-containing designs. Each underwent clinical 1.5-T MRI and conventional CT, and the same Dixon MRI dataset was processed using vendor-provided 2D slice-based and 3D volume-based synthetic CT algorithms.
- Performance was assessed across three complementary domains: bone and soft-tissue geometry using Dice and HD95, CT-number accuracy using two masking approaches, and dosimetric agreement across six 6-MV beam configurations, including individual beams, a four-field arrangement and VMAT.
- The predefined “acceptable” thresholds were demanding and clinically motivated: bone Dice ≥0.8, bone HD95 ≤2 mm, bone mean HU error ≤100 HU, soft-tissue error ≤10 HU, and absolute dose-metric deviation ≤1%. Marginal and unacceptable categories were also defined prospectively.
- Phantoms without an internal skull-equivalent structure essentially failed bone synthesis, with bone Dice ≤0.001 and common-mask bone HU errors ≤−1100 HU. This demonstrates that an external head shape alone is inadequate for meaningful cranial sCT QA.
- Even skull-containing designs remained imperfect. Across both algorithms, bone Dice ranged only 0.24–0.71 and HD95 4.2–24.4 mm. Self-mask bone HU error ranged from −95 to +470 HU, while common-mask error ranged from −198 to −1115 HU.
- Dosimetric agreement was likewise variable: mean absolute Dmean deviation averaged 3.2% for the 2D algorithm and 3.7% for the 3D algorithm, with no significant difference between algorithms (p=0.49). Soft-tissue HU error remained within ±120 HU among the skull-containing phantoms.
- No phantom passed all predefined criteria simultaneously. The investigators concluded that a skull-like structure is necessary but insufficient: MR-compatible, low-susceptibility materials and anatomically realistic tissue properties are also needed, and phantom performance must be judged jointly across geometry, HU fidelity and dose rather than by a single metric.
CLINICAL TAKEAWAY
Departments implementing cranial MR-only planning cannot assume that an ordinary anthropomorphic head phantom is suitable for synthetic CT QA. Current commercial designs may be useful for selected checks, but this study found no off-the-shelf phantom that could act as a complete end-to-end sCT QA standard.