A Deep Learning Force Estimator System for Intracardiac Catheters

Having a real sense of the applied force in catheterization procedures can help surgeons with proper treatment for cardiovascular diseases. Using sensors is not common because of the limitations of catheters and complications related to the safety of patients. In this regard, a sensor free method can be deemed as a safe solution, in which it uses available equipment in the real operation room. In this work, we propose a deep learning method to estimate the contact forces directly from the catheters’ image tip without embedding further sensors. A convolutional neural network extracts the catheter’s deflections through input images and translates them into the corresponding forces. The architecture of the proposed model has been inspired by the ResNet graph so as to perform a regression. The model can make predictions based on the input images without utilizing any feature extraction or preprocessing steps. An experimental setup was designed and implemented to simulate catheter ablation therapy. Evaluation results show that the proposed method is able to elicit a robust model from the given dataset and approximate the force with proper accuracy. Opting RMSE as the preferred performance metric, the model reached 0.028 N and 0.023 N in estimation error in the x and y direction on the test data set, respectively.