Goal: Personalized diabetes prediction
Data
 1074 cases (healthy, diabetes)
 Clinical data: IMC, sex, age.
 Polymorphisms (112): insulin-like growth factor binding protein, CD36 antigen, adiponectin receptor 2, etc.
Method
Dimensionality reduction with PCA and personalized prediction with CBR.
Results
5 principal components, PCA-CBR success: 89%
Medical databases usually consist of a great number of variables. It is therefore tedious for the diagnostic system to cope with all these attributes. Hence, it is important to find methods to reduce the total amount of attributes without the loss of performance. One of the well-known techniques that has been widely applied to reduce the dimensionality is the principal component analysis (PCA). PCA techniques are designed to deal with numerical attributes, but our medical dataset contains many categorical data and alternative methods such as RS-PCA are required. Thus, we use RS-PCA (regular simplex PCA) suitable for discrete data. We hybridize it with a CBR system to support medical diagnosis. Results are quite promising since they allow diagnosis with less computation effort and memory storage.
Healthcare and Medicine

 

CANCER

Goal: Breast cancer prognosis
Data
 347 families (GED files) with gene mutation information.
 554 clinical data of individuals of families (csv files).
Method
 Automatically generate inheritance indicators, use indicators with clinical data to predict cancer risk, case-based reasoning for individual-personal prediction.
Results
8 indicators, 11 inheritance rules, 91 % sucess with mutation inheritance. exitCBR tool.
Breast cancer has been our true collaboration boot in the world of healthcare, challenging us with problems that require tools to guarantee experimental repetition, due to the criticality of the impact of healthcare results. To that end, we develop the eXiTCBR tool based on the case-based methodology, since this is the AI methodology that supports evidence-based medicine. Moreover, it allows the hybridization of other machine learning and reasoning techniques. Nowadays, the tool has growth and there are several versions available. Regarding breast cancer, we developed as part of the tool a new reuse method for case-based reasoning, a genetic algorithm to learn the relevance of clinical variables. A plug-in to manage data files inherited from GED has also been developed, so that family inheritance indexes can be managed at the clinical level. Subgroup discovery techniques have also been applied.
Medicine and healthcare UdG


TAVI

Goal: Clinical workflow support for Transcatheter Aortic Valve Implantation (TAVI)
Data
20 cases H. Clínic (Barcelona).
100 cases cases H. Universtitaire Rennes (France).
Method
Tool retrieves attributes of similar cases according to information available at each stage (task).
Results
Prototype.
Clinical Decision Support System (CDSSs) should form an important part of the field of clinical knowledge management technologies through their capacity This work aims to support the clinical process and use of knowledge, including knowledge maintenance and continuous learning, from diagnosis and investigation through surgery, treatment and long-term care. A workflow-based Clinical Decision Support System (CDSSs) was designed to give case-specific assessment to clinicians during complex surgery or Minimally Invasive Surgery. Following a perioperative workflow, the designed software use the Case-Based Reasoning methodology to retrieve similar past cases from a case base to provide support at any particular point of the process. The graphical user interface allows easy navigation through the whole support progress, from the initial configuration steps to the final results organized as sets of experiments easily visualized in a user-friendly way. As a results, the eXiTCDSS tool was developed.
TAVI

 

STROKE

Goal: Improve stroke classification
Data
 Badisen Stroke Data Bank.
Method
Decision making supported by case-based reasoning, multi-agent system to support team cooperation and assessment.
Results
Prototype.
Acute strokes are medical emergencies that require from expert neurologists in order to detect the illness on the appropriate therapeutic time window. Thanks to the development of new treatments as the rt-Pa treatment mortality rates that have been descending in the last decades. However, the final diagnosis of the patients in often imprecise. That is, patient can have acute stroke as the diagnosis, but the clinical category of it is often unknown. This situation has been detected by the Spanish Association of Neurologists who has set up a repository of cases, Badisen. From its data base, we have designed a  multi-agent case based system with the aim of giving support in the acute stroke diagnoses. An agent in the system keeps information of experiences in a single hospital, maintaining the particular decision criteria employed by the main physician. Agents collaborate in a lack of confidence in the isolated decision problem.
STROKE


ENDODONTICS

Goal: Support learning of endodontics diagnosis
Data
-
Method
E-learning, adaptive profiles.
Collaborative work
We collaborate with Universidad de Rosario, Argentina.
In Endodontic’s, as in other health’s area, learning to take a correct diagnosis is a fundamental activity for the vocational training. In that regard, the University of Rosario implemented the EndoDiag system, an expert system for the student guide in the resolution of a clinical case that he is analyzing. Later on, EndoDiag II includes clinical cases that represents real situations of the professional practice, and adapts the exercise to the student experience. As the complexity of the platform was growth, our work was to explore multi-agent architectures to support the e-learning system.
Endodontics

 

PREMATURE BABIES

Premature babies monitoring at home.
Data
Cooming Soon.
Method
Hybridization of case-based system with other kind of knowledge-based systems, context resasoning, personalization with case-based reasoning, integration with HIS (SNOMED-CT, HL7).
Resultats
Prototype
Recent studies show that premature babies who need a supervision period can shorten this time if they are in a familiar environment. However, complications might arise at any time. This work proposes an approach to monitor babies at home based on mobile phones. Baby sensors send data to a mobile device implementing an intelligent system that provides recommendations and supports the parents in the baby care. The system is composed by a rule-based system and a case-based system, using the knowledge provided by physicians and historical data.  Case-based reasoning enables to personalize the recommendations.  On the other hand, mobile devices enable gathering context information that are also managed by the case-based reasoning engine, improving clinical decision making. In the last step, the mobile device connects (thanks to interoperability standards) with the appropriate hospital information system, such that the medical staff always contrast the assessments derived from the application.
Premature babies

REHABILITATION

Goal. Signal sensor processing to predict rehabilitation period.
Data
 na
Method
Feature characterization, case-based dimensioanlity reduction, personalization with case-based reasoning.
”Results”Prototype[/dlitem
Elderly citizens often suffer from hip illnesses that require surgery (as substituting the hip by a prosthesis). Rehabilitation from such surgery is crucial to recover full walking movements and so guarantee the persons’ quality of life. However, the same rehabilitation programme seems to have different results on patients. Thanks to the advances on sensor technology, today it is possible to monitor the patient’s steps. From the information of the sensor, our challenge is to find patterns that lead to good recovery.
REHABILITATION

 

MEDICAL EQUIPAMENT MAINTENANCE


Goal: Complex equipment failure prediction (e.g. Magnetic Resonance Imaging devices, Computer Tomography scans).
Data
Equipment logs.
Method
Sequence learning + case-based reasoning and probabilistic.
Results
97 % success.

The use of complex medical equipment (Magnetic Resonance Imaging – MRI, Computer Tomography – CT,  Positron Emission Tomography – PET-CT ) has been crucial for patient diagnosis. Such equipment is known to be complex, composed of several pieces and subsystems, all of them subjected to high safety constraints, since they are using sensible technology regarding human health. Thus, if there is a minimal deviation of a given configuration parameter of the equipment component, the equipment sets up an alarm and stops. Nevertheless, when a machine fails, it causes many problems in the clinical service. Thus, a desired situation is to monitor the medical equipment so that failures can be predicted. We use sequence learning to learn failure patterns from even logs produced by the components of the machines. Patterns are used in a case-based reasoning system to predict failures.

On the other hand, we explore a probabilistic approach to create predictive models based on the exploitation of existing maintenance registers for the estimation of failure rates and internal cause-effect relationships to build those models.


Goal: Workflow monitoring for equipment repair. 
Data
Simulated.
Method
Petri nets, agents, auctions, heuristics.
Medical technology is by now an integral part of health care according to consisting generally accepted standards. Their purchase and operation thereby represent an important economic position and both are subject of everyday optimisation attempts. Optimization means that, expensive equipment is repaired quickly, but also cheap but abundant equipment is kept on work. Here, we use Petri nets for workflow modelling and complex event processing (CEP) for workflow monitoring to predict possible delays.

 

AMBULANCE COORDINATION


Urgent patient transport.
Data
44 ambulances, Girona region.

Method
Agent-based system,  Fuzzy filtering regarding ambulance trust; Region coverage.

Results
99,5 % arrivals within 15’.

Emergency transportation on specialized vehicles is needed when a person’s health is in risk of irreparable damage. In rural regions, studies have shown that ambulance teams have different response times, mainly because of the drivers’ expertise. The challenge for the regional authorities is to appropriately coordinate their resources to continuously improve response time, taking into account the resources needed and driver expertise. In response, we developed a multi-agent system that maintain the real distributed organization of resources. To solve the assignment problem (which ambulance agent attends a patient request), we use an auction mechanism and a region coverage algorithm. Driver skills are represented by trust, and ambulance response times are modulated by them.


Programmed patient transport.
Data
700 daily routes, 80 ambulances.
Method
Metaheuristic method.

Results
Product on exploitation (IrisAmb by Lafcarr); Spin-off(Newronia)
The daily programming of patient transport to hospitals and healthcare places due to specific treatments has been carried out for a long time manually, and based on human experts. To automate the system, we develop an optimization engine based on metaheuristics. The systems allows the automatic scheduling of medical transportation services considering a set of constraints as the needs of the user (stretcher or sitting service, individual or collective, equipment, etc.), compliance with current legislation, waiting time and length of service bounded, and the limited availability of resources (ambulance drivers). The system proposes the sequences of services to be done and which ambulance should be carried them out, according to different optimization criteria: kilometres, waiting time of patients, ambulances used.
Ambulance coordination