Poster
# 28
Poster |
|
6th Internet World Congress for Biomedical Sciences |
William Reiser(1), Anita Maria da Rocha Fernandes(2)
(1)Science Computer Course. Universidade do Vale do Itajaí - UNIVALI CTTMar - - Itajaí. Brazil
(2)CTTMar. Universidade do Vale do Itajaí - UNIVALI CTTMar - Science Computer Course - Florianópolis. Brazil
|
 |
|
| |
[New Technology] |
[Health Informatics] |
[Odontology] |
Artificial Neural Networks have been quite used like classifiers (1,2,3) in biomedicine and more recently in dentistry. Among odontology areas that need the use of a classifier aid the decision make, there is dental avulsion diagnosis.
Nowadays, there are few studies based on many bibliographies that can give information about prognostic in a long space of time related to permanent teeth that were avulsioned and reimplanted. So, it´s necessary to give the information needed to the decision make in an accurate and fast way.
By the use of artificial neural networks it is possible to classify the problem (dental avulsion) aiding the dentist in the decision make, because the diagnosis will be done by the classification of many factors, that will be used like the inputs of the neural network.
RNAD is a system to aid the diagnosis of dental avulsion, working like a fast accurate classifier to help the dentist in the decision make.
The general goal of this paper is to present the development of an artificial neural network toaid dental avulsion diagnosis.The specific goals are:
- to study the concepts about artificial neural networks;
- to make available the application of artificial intelligence concepts in health, more specific in dentistry;
- to study MATLAB (5) shell, in order to verify it´s use in neural network development;
- to develop a system that can be able to aid the dentistry in decision make process of Dental Avulsion Diagnosis.
The methodology used in the development of RNAD had the steps bellow:
- acquisiton of information about neural networks;
- study about MATLAB;
- knowledge acquisition about computational aspects that involve the system;
- knowledge acquisition about dental avulsion, in order to determinate the inputs and outputs of the neural network.
The phases that involve the system developed were:
- bibliographical revision: study of concepts about artificial neural networks; study about the tools that were used to develop the system prototype; study of dental avulsion;
- study about MATLAB, its functions and toolboxes;
- knowledge acquisition and representation;
- determination of neural network inputs and outputs;
- system validation.
The system development was composed by: (i) knowledge acquisition; (ii) neural network structuration, trainning and validation; (iii) user interface implementation.
Knowledge Aquisition
During the interview with the dentistry (expert in dental avulsion) were defined the variables that composed the neural network inputs. There were seven variables: conservation environment; time (in minutes); kind of fracture; pacient age; tooth; fracture and kind of dentition. The expert defined nome prognostics that were the desirable outputs, classified in favorable and unfavorables dental reimplant.
According to the expert, the variables were choosen based on the importance in the process.
Neural Network Structuration, Trainning and Tests
According to the neural network nature, the quantity of inputs and outputs were convert in numbers and after they were organized in matrixes, that are the basic element in MATLAB. An example of this convert is:
Variable time: intervals - 0 to 60 = 1
61 to 120 = 2
121 to 360 = 3
361 to 720 = 4
above 721 = 5
After the convert, it was used an Excel worksheet to sort the data.
The algorithm used to develop the neural network was backpropagation. This algorithm was implemented step by step in MATLAB. Although MATLAB has pre-defined functions to develop neural networks, called Toolbox the development team prefer to develop all the algorithm.
In order to choose the best architecture to the neural network it was necessary a lot of tests to stablish some values, such as: number of hidden layer, number of epochs, sum square error, learning rate, momentum. The best set of parameters was: hidden layer = 30; epochs = 2425; sum square error = 0,5; learning rate = 0,1; momentum = 0,0000005.
The neural network trainning needed 296 input patterns (148 favorable diagnosis and 148 unfavorable diagnosis). It was selected in an aleatory way 60 cases to test the neural network input. They correspond to 20% of all data.
User Interface Implementation
MATLAB ia a very good tool to develop neural networks, but its toolbox to develop graphic user interface is poor, so, it was necessary to develop an user interface using Delphi.
The paper wanted to show the applicability of neural netwoks in dentistry. The goal was reached. The results were excelent and the system was tested by dentists, that explained the good aid the system give to them.
To future works we suggest that a test using RBF neural network in this base be done, and the results, compared with backpropagation results.
It will be very good if a test using another artificial intelligence technic like case based reasoning be done to show the applicability of artificial intelligence in dentistry.
|
|
|
|
| |
|
[New Technology] |
[Health Informatics] |
[Odontology] |
