Monte Carlo dosimetry

Monte Carlo (MC) radiation-transport simulation is widely regarded as the reference-standard method for internal dosimetry in nuclear medicine. General-purpose codes such as MCNP6 and dedicated nuclear-medicine codes built on Geant4, principally GATE (Geant4 Application for Tomographic Emission) and GAMOS, simulate the full physics of radioactive decay and the subsequent transport and energy deposition of emitted photons, electrons and (increasingly) alpha particles through a voxelised representation of a patient's anatomy, derived from CT, and their measured or assumed radioactivity distribution, derived from SPECT or PET. Within the MIRD formalism, MC methods are used either to derive tabulated dosimetric quantities (the specific absorbed fraction, and the S-value, or mean absorbed dose per unit accumulated activity) for standard reference phantoms, or to run direct, patient-specific voxel-level simulations.

Validation studies comparing GATE, GAMOS and MCNP6 for reference phantoms and clinical patient datasets have found strong cross-code agreement, typically within about 1-5% for self-organ S-values and within 10% (usually under 5%) for cross-organ absorbed fractions across a range of isotopes including F-18, Lu-177, I-131 and Y-90. Organ-level comparisons between GATE-derived and traditional MIRD/OLINDA S-value dosimetry have shown roughly 2% difference in total organ absorbed dose, though larger discrepancies (tens of per cent) can occur for individual cross-organ pairs. The principal drawback of direct MC simulation is computational cost, historically minutes to hours per patient dataset, which has motivated faster approximations (dose point kernel convolution, voxel S-value convolution) and, more recently, deep-learning surrogates trained on MC-generated dose maps for routine clinical throughput.