Automated Parameters Prediction for Microwave Thermal Ablation of Liver Tumor with Ultrasound Image

Ultrasound-guided microwave thermal ablation (MTA) is one of the major methods for hepatocellular carcinoma (HCC) treatment. However, the current setting of MTA parameters mainly relies on the experience of radiologists and the reference of experimental theory, which easily leads to incomplete or excessive ablation. Therefore, this paper intends to predict the MTA parameters. The MTA parameters include ablation time and power. Considering the difficulty that these parameters do not strictly follow regression or classification rules and the parameters can be flexibly complementary, we propose an energy-guided MTA parameter prediction system based on convolutional autoencoder, back-propagation neural network and clustering. Case feature representations are obtained by fusing tumor image representations and patient clinical information. Finally, the extracted feature representation is used to recommend similar ablation schemes for similar cases. In this study, the Top-K evaluation criterion was used to evaluate the ablation scheme prediction method proposed in this paper. Experimental results show that our method is effective in the prediction of ablation parameters. To our knowledge, this is the first attempt to predict surgical parameter planning for the MTA of HCC. The plan predicted by the model can be used as a preoperative reference plan.