کنفرانس مهندسی برق ایران

صفحه اصلی / سی امین کنفرانس بین المللی مهندسی برق

Transfer learning using deep convolutional neural network for predicting dementia severity

نویسندگان :

Vahid Asayesh¹ Mehdi Dehghani² Majid Torabi Nikjeh³ Sepideh Akhtari khosrowshahi⁴

1- دانشگاه تبریز 2- دانشگاه تبریز 3- دانشگاه شیراز 4- دانشگاه تبریز

کلمات کلیدی :

dementia،electroencephalography،continues Wavelet transform،convolutional neural network،support vector machine

چکیده :

Dementia is a clinical syndrome that includes a group of disorders related to cognitive decline and influence brain functions. Electroencephalography (EEG) is a neurodynamic biomarker that helps in detecting cortical abnormalities and shows good performance in the diagnosis of dementia. Accurate and early diagnosis is important to control dementia progression, for this reason, automatic diagnosis techniques such as using machine learning algorithms have become important in this field. This paper introduces an automatic methodology based on transfer learning with deep convolutional neural networks (CNNs) for the diagnosis severity of dementia. First, EEG signals are converted into time-frequency image using continuous wavelet transform (CWT). Then, the images are applied to the pre-trained CNN. The output of convolutional and pooling layers of this model is used as deep features and are fed into the support vector machine (SVM) classifier. The experiments showed that the F8, Pz, and P4 channels applied to the AlexNet-SVM achieved best results with accuracy of 83.8%, 82.3% and 82.4%, respectively. Also, combination of electrodes in parietal region achieved best results with accuracy of 81.9%.