Microwave ablation temperature maps skip the per-point solve: GAPPTANet predicts the 180-second endpoint in 1.58 ms
Synopsis
Planning microwave ablation needs a 3D temperature field, but solving the coupled electromagnetic-thermal equations point by point is slow. GAPPTANet feeds an analytical specific absorption rate field in as an explicit heat-source prior, uses a geometry-aware adapter to correct applicator-induced near-field approximation errors, and reconstructs global heat diffusion and local high-temperature boundaries with a gated dual-branch decoder. On 27,050 high-fidelity simulations it reaches a full-field root-mean-square error of 0.6001 degrees Celsius and a batch-amortized GPU forward pass of 1.58 ms per sample, outperforming convolutional baselines and three neural-operator architectures under the same test setting.
Interpretation
Microwave ablation temperature prediction can now return a 3D endpoint temperature field in milliseconds instead of re-solving the electromagnetic and heat-conduction equations for each new case. Earlier fast surrogates took the convolutional-network or neural-operator route, whereas GAPPTANet instead supplies an analytical specific absorption rate field to the network as an explicit heat-source prior. Under the same test setting, GAPPTANet reaches a full-field root-mean-square error of 0.6001 degrees Celsius and a mean relative error of 0.5904%, outperforming convolutional baselines and three neural-operator architectures.
The model separates physical prior from geometric correction, so heat source, near-field error and heat diffusion each get their own pathway. The analytical specific absorption rate field supplies the heat-source distribution, the geometry-aware adapter corrects the near-field approximation error caused by the applicator, and the gated dual-branch decoder reconstructs global heat diffusion and local high-temperature boundaries separately. Training and evaluation use 27,050 high-fidelity finite-difference frequency-domain electromagnetic-thermal simulations covering the 180-second endpoint temperature field for a single antenna under a fixed protocol in nonperfused liver tissue.
Perspective
The demonstrated scope is prediction of the 180-second endpoint temperature field for single-antenna microwave ablation under a fixed protocol in nonperfused liver tissue, trained and evaluated on 27,050 finite-difference frequency-domain electromagnetic-thermal simulations, with one limited ex vivo porcine liver experiment as a point-measurement comparison.
The ex vivo porcine liver experiment provides only a pooled mean absolute error of 2.83 plus or minus 1.82 degrees Celsius and is limited in scale; the conclusions address nonperfused liver tissue under a fixed protocol, and behavior under perfusion is not given within the stated scope.