Authors:
Francisco J. Veredas
1
;
Héctor Mesa
1
;
Juan C. Morilla
2
and
Laura Morente
3
Affiliations:
1
Universidad de Málaga, Spain
;
2
Servicio Andaluz de Salud, Spain
;
3
Diputación Provincial de Málaga, Spain
Keyword(s):
Pressure ulcer, Wound, Prediction, Machine learning, Artificial intelligence.
Related
Ontology
Subjects/Areas/Topics:
Artificial Intelligence
;
Biomedical Engineering
;
Cardiovascular Technologies
;
Computing and Telecommunications in Cardiology
;
Data Engineering
;
Enterprise Information Systems
;
Expert Systems
;
Health Engineering and Technology Applications
;
Health Information Systems
;
Healthcare Management Systems
;
Information Systems Analysis and Specification
;
Knowledge Engineering and Ontology Development
;
Knowledge Management
;
Knowledge-Based Systems
;
Medical and Nursing Informatics
;
Ontologies and the Semantic Web
;
Society, e-Business and e-Government
;
Support for Clinical Decision-Making
;
Symbolic Systems
;
Web Information Systems and Technologies
Abstract:
A pressure ulcer is a clinical pathology of localized damage to the skin and underlying tissue caused by pressure, shear or friction. Diagnosis, treatment and care of pressure ulcers are costly for health services. Accurate wound evaluation is a critical task for optimizing the efficacy of treatment and care. Prediction of wound evolution helps the effective management of health resources and planning of pharmacological treatment and health-care decisions. In this paper, different machine learning approaches have been designed and used to predict the evolution of pressure ulcers. These predictive systems are based on local features extracted from wound images which were weekly taken in uncontrolled lighting conditions. The images were automatically segmented by the mean-shift procedure. A group of clinical experts manually classified the segmented regions into five different tissue types, and a set of local descriptors based on area measurements of these tissues was extracted. The on
e-week evolution of two different indicators for pressure ulcer evaluation is predicted: the ratio between granulation and devitalized tissue, and the percentage of wound-bed border consisting of granulation tissue. Of the tens of machine learning approaches and architectures tested in this study, support vector machines, naive Bayes classifiers, neural networks and decision trees achieved the highest accuracy rates in the prediction of the two indicators above, with also acceptable sensitivity and positive predictive value rates. Feature selection significantly reduced the number of input features needed for prediction. Neural networks and decision trees gave the best performance results, and the C4.5 algorithm achieved the highest accuracy rate (∼ 81%) in the prediction of the granulation/devitalized ratio from a small number of input features.
(More)