Research | Projects

MODELING BIO-MEDICAL DECISION THROUGH ENSEMBLE LEARNING - FROM DATA DELUGE TO THE FORMULATION OF PLAUSIBLE SCIENTIFIC HYPOTHESES.

Despite the increasingly growing demand for orthodontic care, that have more and more psychological and social motivations, orthodontic diagnosis, often, appears very difficult to be drawn, resulting, instead, influenced by subjective interpretation of the measured parameters. For instance, treatment planning is a decisive and critical moment for the clinician, especially in case of extraction, which is a non reversible procedure, and that tends to be mainly based on the practitioner's experiences. Moreover, another issue is represented by the diagnostic parameters needed to define the treatment plan: they are very numerous and the relative weight of each parameter for a specific patient is not easy to determine. On the whole, the subjective aspect of orthodontic diagnosis, determines the lack of universality and unanimity in the interpretation of orthodontic data and, consequently, in the treatment selection (Martina, Teti, D’Addona & Iodice, 2006). A referencing framework for orthodontic data evaluation would be desirable and beneficial for a diagnostics treatment selection, particularly in controversial cases, where subjective data interpretation could generate incorrect decisions. As known, decision tree is a classification scheme which generates a tree and a set of rules from a given dataset. They have been widely employed both to represent and to run decision processes (Lopez-Vallverdu, Riano & Bohada, 2012). Considering that medical, and as such orthodontic, decision making are made for various purposes including screening, diagnosing, and treatment prescription/suggestion, the decision problem becomes difficult to visualize and implement. Decision trees represent also an useful and indispensable graphical tool in such settings, as it allows for intuitive understanding about the problem and can aid the decision adoption since trees are interpretable through if-then rules by any orthodontist practitioner, although not trained in using computer applications. Furthermore, the “robustness” of the proposed approach goes far beyond the simplification of the orthodontic decision process, since models extracted from data, through decision trees and ensemble learners, can guarantee the “objectivity” asked by the interpretation of the orthodontic data, as cited above, and, as a consequence, allowing the suggestion of the most appropriate therapy, based on a resolution of a collective of models, previously validated by experts and scholars. Such an approach represents a further attempt, along with the others, towards the foundation of a common framework aiming at reducing, as much as possible, the subjectivity in the interpretation of orthodontic data. References. 1. Martina, R., Teti, R., D’Addona, D., Iodice, G., (2006). Neural Network Based System for Decision Making Support in Orthodontic Extractions, Special Session on ICME, 2nd Int. Virtual Conf. on Intelligent Production Machines and Systems – IPROMS 2006, 3-14 July: 427-432.2. López-Vallverdú, J. A., Riaño, D., & Bohada, J. A. (2012). Improving medical decision trees by combining relevant health-care criteria. Expert Systems with Applications , 39(14), 11782– 11791.