Considerations on the Ontology for Cognitive Surgical Robotics

Cognitive science is a multidisciplinary field that tries to understand cognition at different levels of analysis and with different approaches from psychology to neurosciences to artificial intelligence. A key aspect in this field is knowledge representation, which has evolved with the development of ontologies in computer science to provide a conceptual yet computational model of a particular domain of interest.

With a general definition, an ontology formally represents knowledge as a set of concepts within a domain and the relationships among those concepts. Thus ontologies are always related to a specific domain, which can be either broad or limited, and different types of ontology can be classified depending on the domain they represent. Also, according to van Heijst, ontologies can be grouped as terminological, information or knowledge ontologies.

Cognitive science has been applied to medicine and surgery in particular with ontologies playing a central role. Typical examples of application of cognitive sciences in surgery include knowledge-based and model-based surgery, surgical workflow modeling, computer aided diagnosis (CAD) and context-aware surgical assistance. In particular, an example of knowledge-based system to provide support to clinicians with respect to the drugs they administer to patients in ICUs is reported in.

Jannin et al. proposed a methodological framework for surgical models that includes the definition of a surgical ontology, the development of software for describing surgical procedures based on the ontology, and the analysis of these descriptions to generate knowledge about surgical practice.

Cognition and ontologies also helped in the development of context-aware surgical assistance for example in terms of augmented reality. In order to generate a context-aware assistance it is necessary to recognize the current state of the intervention using intraoperative sensor data and a model of the surgical intervention. In, a visualization is generated, according to the current surgical step, assisting the surgeon to perform the ongoing task, e.g. to warn the surgeon in a dangerous situation or to provide relevant planning information.

Although knowledge is often represented with ontologies, it can also be stored in mathematical models thus model-based algorithms applied to surgery can also be classified within the cognitive surgery paradigm. An example is model-based motion estimation of heart surface for minimally invasive beating heart surgery.

On the other hand, the paradigm of cognitive robotics tries to apply cognitive science to extend capabilities of robots with higher level cognitive functions that involve reasoning, for example, about goals, perception, actions, collaborative task execution and so on. Research on cognitive robotics regards mainly humanoid robots, service robots and multi-robot teams , but also industrial robots.

As concerns ontology for robotics, Hallam proposed an ontology for the science of robotics as opposed to ontologies of robots as objects. The latter describing the physical and technical semantics and properties of individual robots and robot components, while the ontology of the science of robotics encodes the semantics of the meta-level concepts and domains of robotics (e.g. surgical robotics).

Nevertheless in this document we are more interested in a goal-oriented classification of ontologies within the domain of robotics, pointing out the distinction between ontologies representing scholar and encyclopedic knowledge and ontologies as a tool for design and validation of software architectures. While the first one is easier to understand, the second refers to a relatively young research topic in software engineering. Conceptualization has always played a major role in software engineering, e.g. in the early phases of software development, in the definition, use and re-use of software components and as a basis for their integration, and it is recently taking advantage of ontologies.

In particular component-based software more and more often adopts a development process based on ontological modeling of the software components, which can ease standardization, reuse and reconfiguration. In particular for robotic applications, the EU project BRICS (FP7 ICT-231940) is aimed at formalizing the robot development process itself and at providing tools, models, and functional libraries, which allow reducing the development time and enhancing the software quality.

The use of ontology in designing and deploying robot software also allows increasing the safety, e.g. making the system fault tolerant as shown in, where a dynamic architecture adaptation using ontologies is shown.

Cognitive robotic surgery can be thought of as the intersection of cognitive robotics, knowledge-based surgery and context-aware surgical assistance bearing in mind that cognitive robotics is often related to artificial intelligence and to a certain degree of autonomy, which, for surgical robots, is still a controversial topic due to safety reasons. Nevertheless there are research efforts towards autonomous surgical robots, for example in a robot that learns to tie knots, based on recurrent neural networks, is presented.

Another paradigm is context-aware surgical robots. For example sensors information about the surgical environment can be used to control and tune parameters of surgical robots, for collision avoidance or adaptive stiffness control: an ongoing research effort in this direction is the EU project ACTIVE (FP7-ICT-2009-6-270460).

The EuRoSurge Concept
The EuRoSurge project fosters component-based software architecture for surgical robots based on different level of ontological modeling.

The first level is a semantic wiki that starts from an encyclopedic description of particular procedures, which will be extracted from a high quality and collaboratively written paper collection, which we call Robotic Surgepedia following the example of Scholarpedia. Starting from this a second level ontology will be developed, which is more oriented to the software development process, i.e. describing the task and defining the topology of the software architecture. The latter reflects what is also known as the topological model of the architecture, in which interconnection among software modules is defined but specific components are not chosen yet. The actual instantiation of the components is achieved at functional modeling phase. In this phase, which is at the same ontological level of the topological model, ontological descriptions of specific software components instances set up a library that can be automatically used to instantiate the chosen topology in a way that best fits the application requirements.

Eventually, a third level ontology describes the hardware implementation of the system (e.g. cable connections, device parameters etc.) and allows testing and benchmarking and provides warning and contingency actions in case of fault. Figure 1 summarizes the proposed workflow.