Tehran University Medical Journal

Background: Ensuring adequate depth of anesthesia during surgery is essential for anesthesiologists to prevent the occurrence of unwanted alertness during surgery or failure to return to consciousness. Since the purpose of using anesthetics is to affect the central nervous system, brain signal processing such as electroencephalography (EEG) can be used to predict different levels of anesthesia. Anesthesia disrupts the interaction between different regions of the brain, so brain connectivity between different areas can be a key factor in the anesthesia process. This study aims to determine the depth of anesthesia based on the EEG signal using the effective brain connectivity between frontal and temporal regions.
Methods: This study, which is done from April to December 2018 in Tehran, used EEG signals recorded from eight patients undergoing Propofol anesthesia at Waikato Hospital of New Zealand. In this study, effective brain connectivity in the frontal and temporal regions have been extracted by using various Granger causality methods, including directional transfer function, normalized directional transfer function, partial coherence, partial oriented coherence, and imaginary coherence. The extraction of effective connectivity indices in three modes (awake, anesthesia and recovery) was calculated using MATLAB software. The perceptron neural network is then used to automatically classify the anesthetic phases (Awake, Anesthesia, and recovery).

Background: Accurate and early detection of non-alcoholic fatty liver, which is a major cause of chronic diseases is very important and is vital to prevent the complications associated with this disease. Ultrasound of the liver is the most common and widely performed method of diagnosing fatty liver. However, due to the low quality of ultrasound images, the need for an automatic and intelligent classification method based on artificial intelligence methods to accurately detect the amount of liver fat is essential. This paper aims to develop an advanced machine learning model based on texture features to assess liver fat levels based on liver ultrasound images.
Methods: In this analytic study, which is done from April to November 2020 in Tehran, ultrasound images of 55 obese people who have undergone laparoscopic surgery have been used and the histological result of a liver biopsy has been employed as a reference for liver fat. First, 88 texture-based features were extracted from the images using the Gray-Level Co-Occurrence Matrix (GLCM) method. In the next step, using the method of minimum redundancy and maximum correlation, the top features were selected from among 88 features and applied to the classifier input. Finally, using the three classifiers of linear discriminant analysis, support vector machine and AdaBoost, the images were classified into 4 groups based on the amount of liver fat.
Results: The accuracy of the automatic liver fat prediction model from ultrasound images for AdaBoost classification was 92.72%. However, the accuracies obtained for support vector machine and linear discriminant analysis classification were 87.88% and 75.76%, respectively.

Background: Early diagnosis of patients in the early stages of Alzheimer's, known as mild cognitive impairment, is of great importance in the treatment of this disease. If a patient can be diagnosed at this stage, it is possible to treat or delay Alzheimer's disease. Resting-state functional magnetic resonance imaging (fMRI) is very common in the process of diagnosing Alzheimer's disease. In this study, we intend to separate subjects with mild cognitive impairment from healthy control based on fMRI data using brain functional connectivity and graph theory.
Methods: In this article, which was done from April to November 2020 in Tehran, after pre-processing the fMRI data, 116 brain regions were extracted using an Automated Anatomical Labeling atlas. Then, the functional connectivity matrix between the time signals of 116 brain regions was calculated using Pearson correlation and mutual information methods. Using functional connectivity calculations, the brain graph network was formed, followed by thresholding of the brain connectivity network to keep significant and strong edges while eliminating weaker edges that were likely noise. Finally, 11 global features were extracted from the graph network and after performing statistical analyses and selecting optimal features; the classification of 14 healthy individuals and 11 patients with mild cognitive impairment was performed using a support vector machine classifier.

Conclusion: Combining the features of brain graph and functional connectivity by the mutual information method with a machine learning approach, based on fMRI imaging analysis, is very effective in diagnosing mild cognitive impairment in the early stages of Alzheimer’s which consequently allows treating or delaying this disease.

Background: Schizophrenia is a mental disorder that severely affects the perception and relations of individuals. Nowadays, this disease is diagnosed by psychiatrists based on psychiatric tests, which is highly dependent on their experience and knowledge. This study aimed to design a fully automated framework for the diagnosis of schizophrenia from electroencephalogram signals using advanced deep learning algorithms.
Methods: In this analytic study, which is done from April to October 2021 in Tehran, 19-channel electroencephalogram signals from 14 schizophrenia patients and 14 healthy individuals were recorded and pre-processed. Then, the effective connectivity measure using the transfer entropy method is estimated from them and a 19×19 asymmetric connectivity matrix is constructed and represented by a color map as an image. Then, these effective connectivity images are used as inputs to the five pre-trained neural networks of AlexNet, Resnet-50, Shufflenet, Inception, and Xception. Finally, the parameters of these networks are fine-tuned to diagnose schizophrenia patients. All models are fine-tuned based on newly constructed images using the adaptive moment estimation optimizer algorithm and cross-entropy as the loss function. 10-fold cross-validation and subject-independent validation methods are used to evaluate the proposed method.

Background: Early prediction of the outcome situation of COVID-19 patients can decrease mortality risk by assuring efficient resource allocation and treatment planning. This study introduces a very accurate and fast system for the prediction of COVID-19 outcomes using demographic, vital signs, and laboratory blood test data.
Methods: In this analytic study, which is done from May 2020 to June 2021 in Tehran, 41 features of 244 COVID-19 patients were recorded on the first day of admission to the Masih Daneshvari Hospital. These features were categorized into eight different groups, demographic and patient history features, vital signs, and six different groups of laboratory blood tests including complete blood count (CBC), coagulation, kidney, liver, blood gas, and general. In this study, first, the significance of each of the extracted features and then the eight groups of features for prediction of mortality outcomes were considered, separately. Finally, the best combination of different groups of features was assessed. The statistical methods including the area under the receiver operating characteristic curve (AUC-ROC) based on binary Logistic Regression classification algorithm were used for evaluation.

Background: Major Depressive Disorder (MDD) is one of the most prevalent and disabling mental disorders in the world. Due to the life quality decline caused by this disease and its growing nature, timely detection and treatment is of paramount importance. In the present study Electroencephalogram (EEG) signal utilized for the precise detection of MDD using Artificial Intelligence (AI) Methods.
Methods: In this analytic study, which is done in Shahid Beheshti University of medical Sciences in 2023, fifty eight subjects were investigated using an experienced psychiatrist that 30 subjects diagnosed as MDD and 28 determined to be healthy. Nineteen channels EEG signals in resting state with eyes closed situation acquired for five minutes from all of the participants including 36 men and 22 women with the average age of 39.3 years. The EEG signals were preprocessed to remove contaminating signals from brain-originated signals. The EEGLAB package in MATLAB utilized to re-reference channels to the average reference, apply a band-pass filter between 1 and 40 Hz and to remove non-brain components of the signal using Independent Component Analysis (ICA). The cleaned data segmented to the three seconds windows with 50 percent overlapping. These segments were used as the input to the AI models. Deep Learning (DL) models utilized in the present study were EEGNet, ShallowConvNet and DeepConvNet which were developed based on the deep convolutional models for the classification of healthy and MDD brain signals. The main difference between these models laid in the number of specific convolutional layers and the model complexity.

Results: MDD and Healthy signals classification has been done using EEGNet, ShallowConvNet and DeepConvNet models and accuracy of 92.3%, 83.2% and 92.2% were achieved, respectively. Also EEGNet acquired the highest sensitivity of 98.9% and specificity of 79.1%. Conclusion: The detection of MDD patients using EEG signals with high accuracy and generalizability is possible and proposed AI models can be utilized in the clinical settings as assistant tools.