| Sumario: | Objective: his study aims to diagnose diabetes from breath odor using electronic nose device and artificial neural networks (ANN) method in order to support medical diagnosis. Material and Methods: Parameters like blood, urine, sweat and breath odors produced by human body are frequently used for diagnosis of diseases in medicine. Recently, besides lightening the disease diagnosis by parameters of blood and urine analysis, diagnosing some diseases using odors called scientists' attention and many investigations have been conducted about odor. There are about 200-400 different types of common odors in human breath. Additionally, the number of gases sensed and described in breath exceeds 3000. These gases may be recognized utilizing electronic nose technology. In this study, an electronic nose with a code of 41 produced by FIS company was used for odor recognition. Values in human breath were measured utilizing this sensor sequence and they were converted to numerical data. Measurements were done with 10 second intervals and 60 input data were collected for each patient. The data converted to numerical values were collected using two different Artificial Neural Network methods formed in MatLab context. Results: In this study, the methods used for the diagnosis of diabetes from bad breath are Multi Layer Perceptron (MLP), Radial Based Functions (RBF) and Learning Vector Quantization (LVQ) algorithms. Afterwards, performances of all these three algorithms were compared using similar odor data. Conclusion: According to experimental results, all of MLP, RBF and LVQ algorithms are known to provide high recognition probability for classification of various odors. However RBF method was shown to have a better recognition capacity compared to the others.
|
|---|