Accurate and Rapid Prediction of Synthetic Thermal Dose Data for MRI Guided Laser-Induced Thermal Therapy Using a Deep Operator Neural Network
Abstract
Purpose
MRI-guided laser interstitial thermal therapy (LITT) provides minimally invasive option for treating recurrent brain tumors. No accurate and rapid treatment planning methods currently exist for quantitatively predicting patient specific thermal dose. Here we propose a novel method based on deep operator network to address this issue.
Methods
We built the method on a branch-trunk neural network, where the first branch extracts feature from tumor images, the second branch encodes laser parameters (power, duration, and probe orientation), and the trunk models mapping from spatial coordinates to the resulting temperature field. To train and test the model, we generated synthetic datasets (N = 400, 90% for training, 10% for testing) with different noise levels to mimic LITT procedures with two distinct laser probe types (FullFire and SideFire) by a FEM (finite element method) solver. To evaluate model performance in thermal dose prediction (CEM43), we employed Dice Similarity Coefficient (DSC) and Hausdorff Distance (HD). Finally, we evaluated model inference speed.
Results
As data noise increased from 0% to 20%, the method achieved temperature prediction errors from 1.16±0.40% to 2.92±0.25% (FullFire) and 3.79±0.53% to 5.13±1.00% (SideFire). At the noise-free level, the method obtained DSCs of 0.96±0.02, 0.97±0.21, and 0.96±0.21 with corresponding HDs of 0.09±0.08 mm, 0.07±0.07 mm, and 0.06±0.08 mm at 2-, 10-, and 60-minute CEM43 thresholds (FullFire), and DSCs of 0.96±0.02, 0.97±0.21, and 0.96±0.21, with corresponding HDs of 0.20±0.31mm, 0.11±0.30 mm, and 0.15±0.31 mm (SideFire). As the noise level increased, the DSC values showed a decreasing trend, and the HD values increased. The method inferenced a [31 × 31 × 3] temperature map within 0.04±0.05 seconds compared to 24.32±2.12 seconds from the FEM solver.
Conclusion
This study demonstrates the proposed method achieves rapid and accurate prediction of thermal dose with synthetic data. Future work will test it with clinical data.