8.10.7 Computational Human Models
Computational human models may be simple standardized geometric representations or anatomically detailed representations derived from medical imaging. Standardized test phantoms may also be represented computationally for comparison with laboratory measurements. Tissues or phantom materials are assigned frequency-dependent electrical properties and, where thermal calculations are performed, suitable thermal properties. The source position, posture, grounding condition, clothing or accessories, and surrounding objects can also be represented when relevant.
When used with an appropriate numerical method, these models can estimate induced electric field, whole-body and local specific absorption rate, absorbed power density, and the spatial distribution of absorption. They are valuable when these quantities cannot be measured directly in a person and for examining source positions or exposure scenarios that are difficult to reproduce experimentally.
A computational model is a defined representation, not a prediction of the exact response of every individual. Model choice, anatomy, posture, tissue properties, source placement, and spatial averaging can influence the result. Compliance assessments should therefore use the standardized model and procedure required by the applicable method, or justify why another model is suitable.
8.10.8 Model Verification, Validation, And Uncertainty
Verification asks whether the equations have been implemented and solved correctly. It may include comparison with analytical solutions, benchmark problems, reference software, and mesh or time-step convergence studies. Validation asks whether the model adequately represents the real source and exposure scenario for its intended use, commonly through comparison with measurements or established reference data.
The uncertainty evaluation should consider the contributions that can materially affect the result, including source power and phase, antenna geometry and placement, material and tissue properties, mesh or element size, domain boundaries, solver tolerances, environmental representation, model simplifications, and post-processing or averaging. Sensitivity studies can identify which inputs dominate the conclusion.
The assessment record should identify the software and version, numerical method, source and geometry, material data, boundary and excitation conditions, mesh and convergence criteria, output quantity and averaging method, verification and validation evidence, uncertainty, and known limitations. This information is necessary for another competent person to evaluate the credibility and applicability of the result.