Brain-Computer Interface for Medical Robotics

In medical technology, surgical robots are used in human-robot collaboration (HRC) to cognitively relieve surgeons and improve treatment outcomes. A key aspect is training these robots using reinforcement learning (RL), which requires a continuous reward function. Particularly in minimally invasive surgery, such as laparoscopy, deriving such a reward function is complex. The process known as reward shaping demands extensive domain knowledge, the conversion of this knowledge into explicit formulas, and the measurement of all relevant values. 

Brain-Computer Interfaces (BCIs), based on electroencephalography (EEG), offer an innovative way to capture implicit neurophysiological reactions from medical experts and use them for continuous evaluation. These affective evaluations, such as a surgeon's frustration at a robot's faulty action, can be utilized for training. Our project aimed to explore how an EEG-based BCI can be used for a time-continuous reward in robotics.

In realistic HRC application scenarios, robust neurophysiological correlates of emotional-affective reactions (specific EEG patterns) were identified and their suitability for continuous evaluation of a laparoscopic robot was explored. The experimental environment developed for this purpose allowed for targeted simulation of medical actions on a phantom. In two studies involving a total of 25 participants, EEG patterns were analyzed using advanced signal analysis techniques and classified to predict emotional-affective reactions using machine learning algorithms. The results show that deep neural networks, specifically Convolutional Neural Networks (CNNs), are particularly well-suited for extracting appropriate signal patterns for classification. The developed components were integrated in a simulated demonstration, illustrating the enhancement of robot-assisted laparoscopic procedures through the classification of emotional-affective reactions using EEG and deep neural networks.