Explainable AI (XAI)-Based Robot-Assisted Surgical Classification Procedure

Background of the Work Many robot-assisted surgical (RAS) procedures may be arranged in a tree structure. RAS can break down the superstate representing every given surgical job into its constituent states. An important first step toward several automated surgeon-assisting features is the assessment of these discrete states at various time granularities during RAS. Motivation for the Work We provide a deep convolutional neural network (CNN) method called Decay, Transfer, and Compose (DTC) that can estimate both the present super- and fine-grained states simultaneously. DTC can handle anomalies in the dataset by exploring its class limits by means of a class decay process. Statistics from the da Vinci®Xi surgical scheme’s endoscope, robotic arms, and system events are all included into DTC. Contributions in this Chapter When tested on the HERNIA dataset, which was collected during actual robotic inguinal hernia repair surgeries, DTC was found to provide reliable estimations of both the fine-grained public of the procedure. We demonstrate that DTC’s hierarchical structure enhances the state-of-the-art state approximation throughout the full HERNIA-20 RAS operation. We also evaluate the relative importance of different kinds of input data and the architecture of DTC in achieving high precision in estimating the status of the operating room.