Uncertainty Quantification of a Microwave Ablation Simulation with Spatial and Transient Responses
October 3 - 11:00 AM to 11:15 AM
Presented by Gavin Jones, Application Engineer
Part of the
Electromagnetics in the Medical Field Session
This study presents a statistical approach for prediction and uncertainty quantification of a hepatic tumor
Microwave Ablation treatment simulation with a spatial and transient response. The study examines the
variation in the volume of necrotic tissue in a human liver tissue model during Microwave
Ablation/Coagulation therapy due to tissue property variation as a function of wattage and treatment time.
The analysis also assesses the effects of tissue variation on the size, shape, and percentage of necrotic
tissue as a spatially distributed field rather than as a single-value parameter like cross-sectional area or
volume. Data with spatial and functional correlations can often be modeled with fewer training data points
using machine learning techniques. This approach also has the advantage of representing the entire
spatial-transient field, which allows analysis on a greater range of results such as the percentage of
necrotic tissue around an ablation region or areas in which tissue was unaffected. The surrogate models
developed in this study are used to perform sensitivity analysis and uncertainty quantification studies to
assess the variation in volume and shape of necrotic tissues due to variations in human tissue properties
and treatment procedures.
The microwave heating of a cancer tumor model from the COMSOL Multiphysics® Application Library was
utilized in COMSOL Multiphysics® simulation software to simulate a region of hepatic cancer tissue. A
commonly used coaxial microwave ablation antenna, with the corresponding slotted catheter, is positioned in
the center of the cylindrical tissue section. The antenna operates at 2.45 GHz, a frequency widely used in
microwave ablation therapy. A Design of Experiments (DOE) containing both biological and treatment factors
was used to generate proper training data for the surrogate model. The biological factors included
temperature, conductivity, permittivity, and specific heat of cancerous hepatic tissue. The DOE and
surrogate model generation, as well as the sensitivity and uncertainty studies, were performed using SmartUQ
Software Version 5.0.
The surrogate models developed in this study accurately mimic the model behavior of the COMSOL
Multiphysics® simulation. The Continuous Response Emulator, which models summary outputs such as volume of
necrotic tissue, percentage of unaffected tissue, etc., converged with a standardized RMSE of less than 0.15
for all output parameters. The Functional Emulator, which models the spatially and transiently distributed
field of responses for ablation, converged with a standardized RMSE, averaged across all nodes, of less than
0.1. These surrogate models were then used to produce sensitivity results and uncertainty margins due to
expected variation in human tissue properties corresponding to a mock set of liver tumor scenarios.
This study presents a novel methodology for both transient and spatial-transient simulation of tumor
ablation treatment using surrogate modeling. This surrogate model will help ensure a microwave ablation
device is able to produce predictable ablation zones while keeping ablation time to a minimum. The
techniques presented with this case study can be implemented to a wide variety of other medical devices to
deliver similar results and improve treatment predictions on a case-by-case basis for patients.
The Role of Analytics in the Digital Twin
October 3 - 11:30 AM to 11:45 AM
Presented by Gavin Jones, Application Engineer
Part of the
Tools for Optimizing Simulations & Streamlining Workflow Session
Through the use of uncertainty quantification (UQ) and other analytics techniques, this presentation will
introduce attendees to the digital twin process workflow.
For a specific part an authoritative digital truth source can be created. This is a digital, interrogatable
repository of all the accumulated data and knowledge concerning that part. Using efficiently sampled data
from the trade space of a simulation model for the part as well as from physical tests or experimental data
an emulator (or surrogate model) of the simulation may be created and calibrated to match real world
performance. Statistical calibration is one technique that may be used in this process. Statistical
calibration has the advantage over other techniques of accounting for the imperfect nature of all models by
assuming discrepancy between the model being calibrated and the physical data set exists. Understanding the
discrepancy between the simulation model and physical tests or experiments can help identify model form
errors and aid in verification and validation of the simulation.
To illustrate the importance of statistical calibration to creating and running an efficient and accurate
digital twin, an example of a statistical calibration using an emulator built for a model created with
COMSOL Multiphysics® simulation software will be presented.
