Learning Conflicts for C-arm Kinematic Modeling using Artificial Intelligence.

During common surgical tasks related to orthopedic applications, it is necessary to carefully manipulate a mobile C-arm device to achieve the desired position. In this work, we propose the application of learning conflicts analysis to improve the performance of an artificial neural network to compute the inverse kinematics of a C-arm device. Using the forward kinematics equations of a C-arm device (and the respective patient table) a training set for machine learning was generated. However, as an inverse kinematics problem may have multiple solutions, it is likely that training a neural network using forward kinematics data may generate machine learning conflicts. In this sense, we show that it is possible to eliminate those C-arm positions that may represent a learning conflict for the neural network, and thus, improve the accuracy of the model. Finally, we randomly generated a suitable validation set to verify the performance of our proposed model with data different from those used for training.