convolutional neural network

Emotions are time-varying phenomena that are created as a response to stimuli. In order to continuously detect emotions, the response of brain signals and facial expressions to the video stimulus can be used. In this way, a series of stimulating videos are shown to viewers, and at the same time, their brain signals and facial expressions are continuously recorded, and their capacity level (negative to positive emotions) is recorded.

The purpose of this article is to understand human emotions using the analysis of electroencephalography signals. In this article, it is presented to recognize the emotions using the fuzzy convolutional neural network that selects the optimal and effective features from the electroencephalography signal to recognize and recognize the emotional states of different people. In the proposed method, first the electroencephalography signal will be decomposed into different alpha, beta and gamma bands, and then intelligent diagnosis will be performed.

The experimental results show that the relaxation and boring states are better recognized in the alpha band with 94.2% and 78.8% accuracy, respectively. In the gamma band, happiness is recognized better with 92.2% accuracy, and finally in the beta band, fear will be recognized with 92.3% accuracy. Also, the use of fuzzy logic in the proposed method has increased the recognition accuracy in all bands.

It is concluded that the proposed model using CNN has high accuracy in emotion recognition, in addition, the use of fuzzy classification has significantly increased the accuracy of the model.

The brain is one of the most complex and active organs of the body, constantly working and analyzing information and data of the body. Besides, it is responsible for monitoring and regulating the voluntary and involuntary function of other organs of the body. Nevertheless, with aging, people become less mentally active, and the brain gradually gets smaller and smaller, to the point where diseases like Alzheimer's, Parkinson's, and stroke occur.
Alzheimer's disease is a progressive and irreversible brain disease, slowly destroying memory and thinking power, and even depriving a person of the ability to do simple things. It mainly affects those parts of the brain that control memory and language, and over time, it damages more parts of the brain. When this condition happens, more symptoms are seen and the disease worsens. Alzheimer's disease is known as one of the most common diseases of old age, which has a significant impact on people’s everyday lives, causing disability and eventually death. Early detection of Alzheimer's disease can help patients with the disease to slow the progress of the disease by changing their lifestyle. Undoubtedly, if Alzheimer's disease is diagnosed a decade earlier than normal, it will simply be as easy to control. Doctors diagnose Alzheimer's disease with the help of brain MRI, and the use of these images with the help of machine learning methods has become one of the practical solutions for early diagnosing of Alzheimer's disease in recent years.
Although machine learning methods have achieved remarkable accuracy in this area, they still face challenges. Thus, in recent years, to overcome the challenges of machine learning methods, deep learning networks have become a steady foot for medical image classification research. Furthermore, in many studies, it has been used for various applications. The goal of deep learning is to learn the features of high-level hierarchical learning from the features of the lower level, that is, in the elementary layers of simple features such as edges and lines, and in the middle layers the corners, edges, and then higher level features.

In the present study, the convolutional neural network is used, which consists of three layers: convolutional, integration, and fully connective. Correspondingly, a fuzzy layer was used to improve the accuracy of the extracted features between the integration and fully connective layers. In the first step, the image of the brain was given as input to the convolution layer. Then, the fuzzy layer performed the initial distribution of input data in the form of fuzzy clusters. Finally, the last fully connected layers perform the classification and assign the result class label to the corresponding group of clusters. Indeed, given that it is still difficult to distinguish boundaries between classes in the classification of images of complex objects or complex real-world scenes, these classifications are still uncertain or inaccurate, so using a fuzzy layer was expected to improve classification accuracy. Because, unlike classical classification, fuzzy classification means that adjacent classes have a continuous boundary with overlapping regions, and the classified object was determined by the degree to which it belongs to different classes. The purpose of combining CNN and fuzzy logic is for the system to be able to deal with more human-like cognitive uncertainties and to process vague and inaccurate information. In this research, the cross-validationwas based on the k-fold method, where k=10 was considered.

In this study, the experiments use the standard ADNI dataset, which is the most authoritative medical imaging dataset to design and test automatic methods for diagnosing Alzheimer's disease. The MRI dataset included images of 302 people under the age of 75. This group included 211 patients with Alzheimer's disease and 91 healthy individuals. Certain subjects were scanned at different time points and their imaging data were considered separately in the experiments. Evaluating criteria of accuracy, precision, and recall were used to evaluate the proposed model.
The results of the proposed model using the criteria of accuracy, call and F score are reported in table 1.
Table 1. Evaluate the performance of the proposed

Accuracy
Precision
Recall
F-measure
Model


88.35
87.23
86.65
85.43
No use of fuzzy classification


99.61
96.51
95.32
95.61
Proposed

As shown Table 1, the proposed method was compared with and without using fuzzy logic. Accordingly, the results showed that the use of fuzzy logic in all criteria of accuracy, precision, recall, and F-measure, obtained the values of 99.61, 96.51, 95.32, and 95.61, respectively, improved the diagnosis of both groups.
The results of comparing the proposed method were obtained using several other methods such as deep learning, a combination of genetics and support vector machine, a combination of the traditional neural network, perceptron statistical method, unsupervised deep learning, support vector machine, and a combination of shallow networks. As can be seen, the proposed CNN model and fuzzy classification for Alzheimer's diagnosis has achieved 99.61% more satisfaction diagnosis than other related studies to classify Alzheimer's disease and healthy patients.

In conclusion, this study aimed to present an intelligent method of combining CNN and a fuzzy classifier for the early detection of Alzheimer's disease using MRI images. The experimental results were performed on the standard ADNI dataset. The proposed model based on the criteria of accuracy, precision, recall, and F-score, with values of 99.61, 96.51, 95.32, and 95.61, respectively, improved the diagnosis of healthy groups with Alzheimer's disease. It was also observed that the proposed model has higher accuracy than other methods of diagnosing Alzheimer's disease. Therefore, it can be concluded that the use of a fuzzy classifier has significantly increased the model's accuracy.
Ethical Consideration
Compliance with ethical guidelines
The manner of reporting or announcing the research’s results ensures the observance of the material and intellectual rights of the relevant elements (testable, researcher, research, and relevant organization).
Authors’ contributions
Hossein Porgholi: This article was extracted from the master's thesis of the first author who was responsible for project implementation, sample collection, holding analysis sessions, review of the results, and initial writing of the article. Elham Askari: She was the coressponding author and mentor of the implementation stages of the research, and she was also in charge of revising the article.
Funding
The first author funded this study.
Acknowledgments
The authors are grateful to the Islamic Azad University, Fouman, Shaft Branch, Guilan, for supporting and approving this research with code 162416439.
Conflict of Interest
The author declares no conflict of interest.

In this paper, a hybrid brain-computer interface for classification of right and left hand motor imagery using deep learning method is presented to increase accuracy and performance. A hybrid brain-computer interface is designed to achieve a way of communicating between the brain and an external device for patients such as amyotrophic lateral sclerosis. So, the user can control the external device such as a Wheelchair without using any organs of the body and only using brain.

Two electroencephalographic and near-infrared spectroscopy signals were recorded from 29 healthy men and women and pre-processing of the signals was done to eliminate noise. The wavelet transform was used to obtain the scalogram as two-dimensional images for both of the signals, and images were inserted separately from each region of brain and merge region into the pre-trained convolutional neural network to extract feature, classification, and prediction of left and right hand motor imagery.

The results for combination of scalogram images of Frontal-Central and Central-Parietal regions in electroencephalographic signal reached 88%, for Near infrared light spectroscopy reached 85% and for merge of two scalogram images reached 90%.

The combination of scalogram images and the deep learning method used in this study reached significant improvement in the prediction accuracy of right and left hand motor imagery for wheelchair motion control.