نشریه مهندسی پزشکی زیستی
سال هفدهم شماره 2 (تابستان 1402)

Neuronal synchronization as a significant cognitive phenomenon of the human brain, has attracted the interest of neuroscience researchers in recent years. This phenomenon is generally investigated in discrete and continuous neuronal models or experimentally recorded signals of the brain. In this study, for the first time, we investigate the weight synchronization instead of neuronal synchrony, in the training step of the artificial feedforward neural networks. The findings of the study show that the generalized weight synchronization occurs both during the training mode and in the non-training mode. Furthermore, as the training is completed, the synchronization increases between the weights. In this study, a new method is introduced in order to detect synchrony patterns using signal derivative and hierarchical clustering. We have also presented a criterion to quantify weight synchronization in different layers of the neural network. Accordingly, the results demonstrate that the lower layers of the network have a significantly higher level of weight synchrony than the upper layers.

Breast cancer stands as the most prevalent form of cancer among women, with over 80% of early-stage breast abnormalities being benign. Timely detection is therefore crucial for prompt intervention. Ultrasound Radio Frequency (US RF) signals represent a non-invasive, and real-time screening method for breast cancer, offering advantages in tissue differentiation and cost-effectiveness without requiring additional equipment. This research aims to present an intelligent approach for the classification of benign, suspicious, and malignant breast lesions based on effective features extracted from the time series. The dataset, registered as USRFTS, comprises 170 instances recorded from 88 patients. The proposed methodology encompasses four key phases: pre-processing, feature extraction, feature selection, and classification. In the pre-processing phase, B-mode images are reconstructed from US RF time series, and a radiologist manually selects the Region of Interest (ROI) in each image. Subsequently, diverse features in the time and frequency domains are extracted from each ROI during the feature extraction stage. The ant colony method is employed for the selection of impactful features. The dataset is then subjected to evaluation using classifiers such as Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Decision Tree (DT), Linear Discriminant Analysis (LDA), and a reference classification method (RCM). The results demonstrate a maximum classification accuracy of 94.95% for two classes and 93.33% for three classes

Despite the extensive progress in the field of biomechanics of human gait, a suitable gait model with the ability to simulate the control system of the human brain has not yet been presented, especially in 3D mode. The importance of the issue increases when the simulation of human walking is one of the main requirements of designers of biomechanical equipment such as artificial organs, wearable robots and humanoid robots. Regarding the constraints and complexities of previous studies, in this research, a forward dynamic 3D model of gait based on sliding mode controller (SMC) is presented, which simulates the walking behavior of healthy individual on the ground in different movement phases. One of the strengths of this research is the comprehensive and analytical review of 3D rotation consequences of the joints coordinate systems, which is done with 11 DOF inverse dynamic model. Based on the obtained results, the SMC controller is well able to produce stable 3D human gait. Also, in 3D gait analysis, the Cardan rotation sequence is not suitable and YXZ order should be used. This outcome is a very useful result for 3D motion generation for human like walking pattern. The results of this study can be used in the design of humanoid robots, active and passive prostheses. Also, the presented model can simulate the walking of an amputee with a prosthesis and the role of the controller in the path, which is very important and beneficial in terms of rehabilitation.

Esophageal mobility disorders are a type of digestive system problem characterized by abnormal bolus movement in the esophagus. The standard diagnostic method for these kinds of disorders is High-Resolution Manometry (HRM). Despite the availability of guidelines like “Chicago” for the analysis of HRM results, diagnosis is still a challenging task that relies on the physician's skills or requires additional assessment modalities. Additionally, it is typical for esophageal mobility disorders to co-occur in one person, leading to a more complex situation for problem identification.The current study focuses on cases who suffering from more than one disorder simultaneously. Then the problem of disorder identification can be interpreted as a multi-label classification problem. Consequently, the fuzzy classifier architecture that was previously introduced for automatic single-disorder diagnosis by the authors is modified. The presented classifier in this paper not only learns the input space from the samples but also utilizes the co-morbidity of disorders to enhance the prediction results. The outcomes show that adding this information to the learning procedure of the base classifier enhances its performance and generates a new fuzzy classifier that overcomes other multi-label classifiers. The presented method is able to differentiate esophageal mobility disorders with an average Hamming loss of 0.18 ± 0.08 which is better than other competitor methods.

Currently, the rapid growth of the beauty industry, along with the development of intelligent models based on machine learning algorithms, has led to an increase in extensive research in this field. Rhinoplasty is one of the most common and demanding facial cosmetic surgeries because the nose is the most prominence element of the face, which has a great impact on its attractiveness. The purpose of this article is to present a machine learning-based framework for predicting nasal aesthetic evaluation. In this article, a set of geometric parameters of the nose relative to the entire face are given as input and human rating as output to the popular machine learning regression algorithms. An ablation study was then carried out to examine the influence of facial shape, skin color, and texture on the beauty of the nose. Multilayer perceptron classification, K-means clustering, and grey level co-occurrence matrix were used to extract facial shape, skin color, and texture. The results show that the model based on geometric parameters has a moderate correlation with human rating, and by adding each subset of the features of face shape, color, and skin texture, the correlation of the obtained model increases until a high degree of correlation is achieved. The results also show that the random forest algorithm has the best performance among other algorithms based on the evaluation criteria of absolute mean error, root mean square error, and Pearson correlation. The results of this study show that the proposed framework can be helpful in determining the beauty of the nose.

Hypertension is the leading cause of death worldwide. Continuous blood pressure (BP) measurement is crucial for the elderly and people with myocardial infarction, cardiovascular disease, kidney disease and gestational hypertension. Cuff-based blood pressure Holters are the most common method for continuous blood pressure measurement, but due to the use of an inflatable cuff, they often cause discomfort, particularly during sleep. A solution to such problems is the optical measurement of blood pressure using the photoplethysmogram (PPG) signal. This paper introduces a transfer deep learning framework for estimating systolic BP (SBP) and diastolic BP (DBP) using a single PPG signal. The proposed framework consists of three main parts: 1) downsampling by a factor of 4 aimed at reducing model complexity, 2) designing a pre-trained model including CNN and BiLSTM layers, and 3) personalizing the pre-trained model for each patient through transfer learning. We carry out Bland-Altman and correlation analysis to compare our method to the invasive arterial catheter (the gold-standard BP measurement method). Our model was validated on a wide range of BP signals acquired from 100 patients in MIMIC-III database. Results showed that the error and Pearson correlation coefficient of our model are 0.14±7.38 mmHg (mean ± standard deviation) and 0.95 for SBP, and 0.00±4.67 mmHg and 0.92 for DBP. The proposed method satisfies the requirements the AAMI and IEEE-1708a standard and receives a grade A according to the BHS standard. This research has shed light on long-term BP monitoring and the prevention of cardiovascular events.

Lung cancer is caused by the irregular and uncontrolled growth of cancer cells in the lung tissue. Cancer cells find the ability to divide and increase in an irregular and uncoordinated manner. The result of this proliferation is the formation of a cancerous mass in the lung. Lung cancer can start in different parts of the lung, such as the bronchi (the air tubes that connect to the lungs) or non-bronchial tissues, and quickly spread to other parts of the body. The precise understanding of the mechanism of lung cancer is still a complex issue and many researches are being conducted in this field. However, early diagnosis has an important impact on the disease treatment process. Therefore, in this research, the diagnosis and classification of this disease is discussed with the help of deep learning and learning transfer. In this regard, the pre-trained Alexnet network has been selected. During the process of transfer learning, the network for lung cancer detection is set on IQ-OTH/NCCD data in three categories: normal, benign and malignant. For this purpose, the last all-connection layer of the Alexnet network is removed and replaced by a new all-connection layer corresponding to the number of layers in the dataset. The classification accuracy of the proposed method on the IQ-OTH/NCCD dataset is reported to be 93%.