Clinical decision system for GIST cancer

In my talk I will present to you the approach we have taken to implement the Cdss application and the application itself in a demo. 

The objective of the project were to build a clinical decision support system for oncological treatment and given our background, we decided to use semantic technologies to enhance the data consistency and to simplify the future adaptation of the application to other use cases. The main challenges that we faced at the beginning were: how to formally describe the knowledge related to the clinical history of a patient, how to model forms based on that information and how to reason efficiently over the patient's information?

In a first part of my presentation, I will show you that after a first period of close collaboration with the knowledge expert, we modelled the knowledge related to the patient clinical history using Ontorion Controlled Natural Language (OCNL) that is a natural language way of expressing OWL2 ontologies. Furthermore we organized this knowledge in stages (or forms) depending on the various steps of the patient treatment. We also defined how and then transitions between the forms should happen by using SWRL rules.

Using this formalization and our knowledge management server Ontorion, we implemented a web application that is using reasoning to create the forms, to decide which is the next form and to show recommendations for the physician. The form is rendered based on the knowledge modelled in the ontology, the next form is decided by computing the result of SWRL rules and the recommendations are reasoned by using the patient history knowledge. In the presentation, I will show how we formalized the knowledge need for the web application and how we used Integrity Constraints, our own extension to the standards to model the forms and its content.

In a second part of my presentation, I will present to you the obstacles that we faced and the ways that we found to overcome these problems. The main obstacles where: to interact effectively with the domain expert and to explain to her how to model knowledge and to solve various kind of reasoning problems related to the limitation of reasoners. I will show you that we found a way to interact with the domain expert thanks to the OCNL by sharing sentences that the domain expert can understand. I will also show that reasoning problems have been solved by developing innovative techniques to limit the size of the knowledge that we are reasoning on and by modelling knowledge that is within OWL-RL+ (our extension of OWL-RL reasoning profile).

In a third part of my presentation I would like to present the next steps that we are taking to make the final deployment at MCMCC site and the extensions we are currently making to the application in order to support more use cases and to deal with data coming from the oncological centre.

Finally I will present to you the web application by making a short demo showing you a live version and showing all the features of this application.