Toward Process Controlled Medical Robotic System

TL;DR

The system design, software implementation, and example application to Robot-Assisted Transcranial Magnetic Stimulation and robot-assisted femoroplasty are discussed, with assessments using performance in terms of robotic tool placement and response to failure injections.

Abstract

Medical errors, defined as unintended acts either of omission or commission that cause the failure of medical actions, are the third leading cause of death in the United States. The application of autonomy and robotics can alleviate some causes of medical errors by improving accuracy and providing means to follow planned procedures. However, for the robotic applications to improve safety, they must maintain constant operating conditions in the presence of disturbances, and provide reliable measurements, evaluation, and control for each state of the procedure. This article addresses the need for process control in medical robotic systems, and proposes a standardized design cycle toward its automation. Monitoring and controlling the changing conditions in a medical or surgical environment necessitates a clear definition of workflows and their procedural dependencies. We propose integrating process control into medical robotic workflows to identify change in states of the system and environment, possible operations, and transitions to new states. Therefore, the system translates clinician experiences and procedure workflows into machine-interpretable languages. The design cycle using hFSM formulation can be a deterministic process, which opens up possibilities for higher-level automation in medical robotics. Shown in our work, with a standardized design cycle and software paradigm, we pave the way toward controlled workflows that can be automatically generated. Additionally, a modular design for a robotic system architecture that integrates hFSM can provide easy software and hardware integration. This article discusses the system design, software implementation, and example application to Robot-Assisted Transcranial Magnetic Stimulation and robot-assisted femoroplasty. We provide assessments using performance in terms of robotic tool placement and response to failure injections.

Figures (14)

• Figure 1: Design cycle of a PCMRS system. FSM design, SB (State Branch) identification, hFSM design, and hFSM classes implementation are introduced in Section \ref{['sec:hFSM']}. The red arrow indicates the conversion from hierarchical FSM to software implementation, which is deterministic and thus can be automated. The conversion is a process that translates hFSM into codes. D-SFO (DispatcherState/Flag/Operation paradigm) implementation is introduced in Section \ref{['sec:d-sfo']}. In future, the step 1 can be an automated process as shown in ongoing research in generative models strong2023chatbot. Subsequently the design cycle can be fully automated.
• Figure 2: PCMRS system diagram. This figure illustrates the organization and communications between software modules in a PCMRS system with minimal extensions used for RATMS. Boxes in blue with bold text are modules implemented using ROS packages in C++. Boxes in green are modules implemented using non-ROS software. Specifically, 3D Slicer module is the stand-alone application with customized Python scripting, Unity module is the Unity platform with C# scripts, and the Kuka iiwa robot interface is developed in C++ libraries. 3D Slicer is an example of planning and visualization module, and can be replaced by any software with TCP/UDP interfaces. Boxes with dashed lines denote classes with weak ownership that are associated by aggregation (Unified Modeling Language terminology) with those in boxes with solid lines. Green arrows denote connections in UDP, and blue arrows are in ROS topics. Note that each blue box is a ROS package containing several ROS nodes, and each blue arrow may indicate several ROS topics.
• Figure 3: The RATMS setup. In this procedure, Electromyography (EMG) is used to measure motor evoked potential (MEP) in response to TMS pulses delivered at the motor cortex. The target stimulation point is at the hand knob within the primary motor cortex lemon2008descending. The example EMG waveforms with TMS are shown from 4 different subjects. The EMG signal is measured with electrodes attached to the first dorsal interosseous (FDI) in the hand contralateral to the side of TMS application (i.e, left hand with TMS at the right motor cortex).
• Figure 4: Uncontrolled workflow for an RATMS system liu2022inside. The directions of the arrows in the figure represent data flow.
• Figure 5: Fig. a. An example FSM containing 3 possible states and 4 possible operations. The system starts at State 1 and transitions to State 3 by completing Operation B. Successful Operation A and Operation C cause alternative state switching between State 2 and State 3, following the arrows connecting the states. At State 2, the successful completion of Operation D causes a transition back to State 1. The system will forbid any other operations not depicted in the diagram, and any failed execution of the depicted operations will not lead to a state transition. Fig. b. A simple hierarchical FSM. The State 2 in a. is expanded to be another FSM.

Theorems & Definitions (5)

• Definition 1
• Definition 2
• Definition 3
• Definition 4
• Definition 5