S. Fazeli

Publications (3)0 Total impact

• Conference Proceeding: An Adaptive Neuro-Fuzzy Inference System for Diagnosis of Aphasia

  S. Fazeli, M. Naghibolhosseini, F. Bahrami
  [show abstract] [hide abstract]
  ABSTRACT: Aphasia is a language disability that has several subdivisions such as Anomic, Broca, Global, and Wernicke. Some reasons such as dissension in description of aphasia and its symptoms, large number of test items which are not quite accurate, linguistic ambiguity and uncertainty as well as typical complexities of medical diagnosis cause accurate diagnosis of aphasia to be a particularly difficult and error prone medical task. To address the diagnosis of the four mentioned common types of aphasia more efficiently, an adaptive Neuro-Fuzzy inference system (ANFIS) is proposed. This structure models the nonlinear relation between aphasia symptoms and resulting test scores as well as the degree of fuzzy belonging of the symptoms to all four major aphasia simultaneously. The proposed method in this paper is compared with a hierarchical fuzzy rule-based structure and a back propagating feed-forward neural network. Our method reaches to a maximum accuracy of 94.6% in 50 trials while the best result for other methods previously used is 91.6%. This method not only diagnoses the four types of aphasia accurately, but also it is efficient for other medical diagnostic applications.
  Bioinformatics and Biomedical Engineering, 2008. ICBBE 2008. The 2nd International Conference on; 06/2008
• Conference Proceeding: Fuzzy Clustering of Transient Evoked OtoAcoustic Emission Signals Using Gustafson Kessel Algorithm

  M. Naghibolhosseini, A.H. Jafari, S. Fazeli, C. Lucas
  [show abstract] [hide abstract]
  ABSTRACT: Healthy ears reflect OtoAcoustic Emission (OAE) signals in response to acoustic stimuli that can be recorded in the external ear canal. We can use the existence of this signal to check the health condition of hearing. Early diagnosis and treatment of hearing abnormalities of infants can save them from losing their hearing. In this paper we use a dataset which consists of OAE signals of subjects with normal and abnormal hearing. First fundamental features which are responsible for the existence of OAE are extracted from unlabeled data, and then Gustafson Kessel clustering Algorithm is applied to the feature space to classify them in two clusters. After that we allocate each labeled data to these clusters based on their distance from cluster centers. Finally we label each cluster based on majority voting rule for their nearest neighbors which are labeled. The best result is 96.6% of accuracy that is achieved from executing the algorithm in 20 trials. Checking the health condition of hearing using OAE signals for experts usually is an error-prone diagnosis, thus to reach an accurate diagnosis, additional tests such as audiometry and tympanometry are commonly used. These tests are time consuming and costly. Reducing these disadvantages leads us to automate diagnosis as a supporting tool to help the experts.
  Bioinformatics and Biomedical Engineering, 2008. ICBBE 2008. The 2nd International Conference on; 06/2008
• Article: Fuzzy clustering of transient evoked OtoAcoustic Emission signals based on K-nearest neighbours rule

  M. N. Hosseini, A. H. Jafari, S. Fazeli, S. Mahmoudian
  [show abstract] [hide abstract]
  ABSTRACT: Every healthy ear reflects a signal (OtoAcoustic Emission, OAE, signals) in response to a stimulus, so we can check the health condition of hearing based on the existence of this signal. If the hearing abnormality of neonates is diagnosed in the early month of their birth, with early treatment we can save them from losing their hearing. In this work we use a data set recorded from subjects with normal and abnormal hearing. First, basic features responsible for the existence of OAE are extracted from unlabeled data, and then KNN-FCM Algorithm is applied to the feature space to classify them in to two clusters. After that we illustrate that each labelled data belongs to which of the clusters based on their distance from cluster centers. Finally we label each cluster applying majority voting rule for those nearest neighbors that are labeled. The best result is 96.6% of accuracy that is achieved from executing the algorithm in 20 trials. Checking the health condition of hearing using OAE signals for experts usually is an error-prone diagnosis, thus to reach an accurate diagnosis, additional tests are commonly used. These tests maybe time-consuming and costly, reducing these disadvantages leads us to this research to automate diagnosis as a supporting tool to help the experts.