Informatics Publishing Limited

Amer A. Sallam ¹, Amin Saif ², Mogeeb A. Saeed ¹, Siham A. Mohammed ³

Affiliations
1 Department of Computer Networks & Distributed Systems, Taiz University, Taiz, YE
2 Department of Mechatronics Engineering & Robotics, Taiz University, Taiz, YE
3 Department of Information Technology, Faculty of Engineering & IT, Taiz University, Taiz, YE

Abstract

Authentication with textual password has several limitations: passwords have low entropy in practice, are often difficult to remember, are vulnerable "shoulder surfing". Biometric system does not meet requirement as well. It relies upon unchanging features that have a lifetime as long as the individual. To avoid this limitation, we start to authenticate with thinking pass thought. User performs one mental task such as thinking of a word or phrase. In this study, Electroencephalography (EEG) was used as method for monitoring andrecording the electrical activity of the brain. These signals can be captured and processed to get the useful information that can be used in pas-thoughts authentication system. Suitable analysis is essential for EEG to differentiate between best and worst tasks used for authentication. This study focuses on usefulness of EEG signal to identify best tasks suitable for the pass-thoughts authentication system. Artificial neural network (ANN) is used to train the data set. Then tests are conducted on the testing data of EEG signal to identify best and worst tasks suitable for authentication. Finally, the system performance was evaluated by computing the accuracy and thereforepromising results were obtained.

Keywords

Electroencephalography, Artificial Neural Network, Discrete Wavelet Transform.

Full Text

   

Amin Saif ¹, Yakoop Razzaz Hamoud Qasim ¹, Habeb Abdulkhaleq Mohammed Hassan Al-Sam ¹, Osamah Abdo Farhan Ali ¹, Abdulelah Abdulkhaleq Mohammed Hassan ¹

Affiliations
1 Department of Mechatronics and Robotics Engineering, Taiz University, YE

Abstract

Lung cancer is the leading cancer cause of death and it's survival rate is very small. An early diagnosis is a good solution to increase the survival rate for lung cancer. To diagnose lung cancer using deep learning we present a convolutional neural network to diagnose three types of lung cancer (Adenocarcinoma, Benign and Squamous) based on histopathological images. The proposed model consists of a main path and three sub-paths. The main path works to extract the small features and creates feature maps at low-level. As for the sub-paths is responsible for transferring the medium and high levels feature maps to fully connected layers to complete the classification process, also the VGG16 was prepared to compare it with the performance of the proposed. After training the models and testing them on 1500 images, we obtained an overall accuracy of 98.53% for the proposed model and 96.67% for the VGG16 model. The proposed model achieved a sensitivity of 97.4%, 99.6% and 98.6% for Adenocarcinoma, Benign and Squamous respectively. We can say that it is not always necessary for the used to be very deep to diagnose histopathological images, and the most important thing is to create a sufficient number of feature maps at different levels.

Keywords

Deep Learning, Convolutional Neural Networks, Lung Cancer Detection, Transfer Learning.

Full Text