Artificial Intelligence for Clinical Decision Support
PhD project information
This PhD work will develop a new tool able to support clinical decisions based on evidences to be activated in multiple clinical contexts and workflows.
The system will leverage the large set of information available in repositories and multiple clinical solutions to create a multidimensional view of patient reality and, starting from this model, will be able to infer possible consequences and suggest reactions.
Inferential models will be based on predefined pathways as well as statistically derived evolution paths created by AI, machine learning and process mining algorithms.
- Industrial partner: Exprivia SpA
- Academic/research partner: Fondazione Bruno Kessler
- Number of available PhD positions: 1
- Duration: 3 years
- Innovation Focus Areas(s): Digital Wellbeing
- DTC location: Trento
- This PhD will be funded by EIT Digital and Exprivia SpA
PhD thesis motivation and innovation valorisation
The adoption of digital tools in health care is driven by different requirements, including the need to simplify collaboration (especially with administrative instruments), create historical logs of every action for legal purposes and, last but not least, to support clinical decisions and workflow.
The actual use of electronic clinical information to evaluate human decisions against predefined protocols or against statistically known evolution patterns, however, is still largely underexploited.
A few software solutions able to support human decisions in specific clinical domains exist on the market, but no currently available market solution is able to extensively and transparently support clinical decisions and pathways taking advance of the enormous amount of information available in the healthcare systems.
This PhD thesis will study, design and implement an intelligent component, integrated in a complex integrated clinical products solution, able to monitor every useful information to obtain two important results:
- The first one is a real-time match of patient conditions against configured pathways models able to recommend (and synchronize) actions introducing a different perspective in applications (M2M) and humans (M2H) collaboration.
- The second one is a fluid and continuous enrichment of statistical models allowing an internally managed AI to identify clinical conditions and propose automatic advice, such as recommending protocols, proposing clinical actions with their statistical outcomes, identifying statistically relevant adverse events, etc.
The PhD thesis will develop a number of different solutions, to be integrated in a complex medical software system. The key problems to be solved are the following:
- Introduce new and more specific ways to represent patient condition as the result of selected information
- Introduce a new and more specific way to represent multidimensional evolutions and protocols for managing clinical conditions
- Integrate Artificial Intelligence algorithms with the abovementioned components in order to constantly apply machine learning tools to support medical decisions
- Develop an effective human interface for the previously described integrated components
These innovations will result in a framework able to deliver a different experience in healthcare practice and workflow, introducing a seamless continuous support to clinical human decisions, through history based alerting and machine learning based suggestions.
Expected academic outcomes
The approach proposed in the thesis relies on the efficacious combination of AI techniques: namely, knowledge representation and reasoning and machine learning, with business process representation and execution. This topic is state of the art of the current research to design and implement innovative decision support tools to effectively support flexible care processes.
Firstly, the application of these techniques to healthcare pathways is of great interest because of the peculiarities of the domain, characterized by i) need of flexibility in the care pathway execution, given the that this has to deal with a potential great number of exceptions (e.g., onset of comorbidities), ii) availability of a great amount of clinical data about the patient and about similar cases, which should be exploited by the system to learn from previous experience to suggest the best choice to the doctor, on one hand, and to possibly modify the sequence of future actions or the constraints on actions in the model, on the other and iii) the requirements to let the final decision to doctor, who must have all the information on the patient easy to read and interpret. Secondly, dealing with and solving these problems in a specific complex domain can in turn advance the state of the art in the research field of techniques to combine logic reasoning, machine/deep learning and process representation and execution to build smart AI-based tools. Thirdly, the implementation of this approach in a component designed to be used in the everyday practices can open the way for the evaluation of these techniques in real clinical settings, thus giving substantial feedback to the academic research. The academic outcomes will include papers publications and participation in relevant academic events.
 Claudia d'Amato, Volha Bryl, Luciano Serafini: Semantic Knowledge Discovery and Data-Driven Logical Reasoning from Heterogeneous Data Sources. URSW (LNCS Vol.) 2014: 163-183
 Alessandro Artale, Diego Calvanese, Marco Montali, Wil M. P. van der Aalst: Enriching Data Models with Behavioral Constraints. Ontology Makes Sense 2019: 257-277.
 McWilliams CJ, Lawson DJ, Santos-Rodriguez R, Gilchrist ID, Champneys A, Gould TH, Thomas MJ, Bourdeaux CP. Towards a decision support tool for intensive care discharge: machine learning algorithm development using electronic healthcare data from MIMIC-III and Bristol, UK. BMJ Open. 2019 Mar 7;9(3):e025925.
 Prahs P, Radeck V, Mayer C, Cvetkov Y, Cvetkova N, Helbig H, Märker D. OCT-based deep learning algorithm for the evaluation of treatment indication with anti-vascular endothelial growth factor medications. Graefes Arch Clin Exp Ophthalmol. 2018 Jan;256(1):91-98.
Abajian A, Murali N, Savic LJ, Laage-Gaupp FM, Nezami N,  Duncan JS, Schlachter T, Lin M, Geschwind JF, Chapiro J.
Predicting Treatment Response to Image-Guided Therapies Using Machine Learning: An Example for Trans-Arterial Treatment of Hepatocellular Carcinoma. J Vis Exp. 2018 Oct 10;(140). doi: 10.3791/58382.
Concrete innovations expected as the outcome of the proposal
The first version of the Decision Support System core engine, enabling symbolic and machine learning based suggestions and alerts to be generated from clinical data, is expected to be available at the end of the second PhD year. Industrialization and MDR certification of this version of the tool is planned to take place in third year, leading to a new tool to be proposed as:
Improvement to existing Products: the tool will be integrated in the e4cure suite, the Exprivia Clinical Information System, as an add on module to be proposed to existing and new customers.
New product: Exprivia will propose this tool as an add on to any existing Clinical Information System and/or Electronic Medical Record.
Expected impact of the PhD outcomes with respect to their business line
Exprivia e4cure Clinical Information System is the Exprivia product for the management of all healthcare and clinical processes.
This product will greatly benefit from this framework, which will introduce for clinicians a very innovative, useful and efficient support in care management.
With this support, clinicians will be assisted with alerting when late on decisions, suggestions on possible strategy and forecast about statistically known outcomes, evidence of variance of decisions against defined protocols.
The impact will be twofold:
- the envisaged tool will be proposed to all existing customers (more than 200 hospitals and 5 main regions using Exprivia healthcare products)
- the tool will greatly increase the Exprivia products competitiveness, thanks to a unique set of advanced, AI based features
PhD thesis time-line and milestones
|M6||Architecture design and selection of methods and approaches to pathways representation, rule and alert engine, machine learning tools|
|M12||First system prototype implementing symbolic pathway/rule components and basic APIs for integration with EMR systems|
|M18||Report on first experimental application of machine learning algorithms to selected clinical data extracted from historical databases. Discussion of results and selection of machine learning algorithms to be integrated in the system.|
|M24||Second prototype of the system, integrating symbolic and machine learning engines. Refinement of APIs for activation of symbolic and machine learning algorithms.|
|M30||Second application of machine learning algorithms to clinical data. Production of at least two machine learning models applied to disease evolution prediction and best therapy selection based on patient outcomes. Report on machine learning models results, including evaluation of sensitivity and specificity.|
|M36||Refined complete prototype integrating symbolic and machine learning models in complex clinical pathways, integrated with the EMR. Report on final system and use cases employing the CDSS rule engine for alerts (such as infection alerts), and prediction of disease evolution/best therapy.|
International mobility plan
The mobility destination(s) will be chosen with the objective of combining the presence of Exprivia branches (Germany, France, Spain, Netherlands, Belgium, Poland and UK) and academic centers collaborating with DISI.
Will the PhD Student do the Business Development Experience at Industrial Partner premises?