Loading...
Search
| Friend's email | |
| Your name | |
| Your email | |
| enter code | |
This page was sent successfuly
Digital Object List
Bookmark
Review List
159 viewed
تحلیل داده های حسگری به منظور کنترل فرد محور پمپ های کمک تنفسی در عارضه آپنه
اکبری، علی Akbari, Ali
Analysis of Sensory Data in Apnea Patients to Aid in Heterogeneous Control of Breathing Aid Pumps
Akbari, Ali | 2024
159
Viewed
- Type of Document: M.Sc. Thesis
- Language: Farsi
- Document No: 57634 (05)
- University: Sharif University of Technology
- Department: Electrical Engineering
- Advisor(s): Jahed, Mehran; Hossein Khalaj, Babak
- Abstract:
- Sleep apnea is a hidden threat to public health, with 80% of individuals being unaware of their condition. Various methods and approaches exist for examining this disorder. In the diagnostic approach to this disease, the data from sleep clinics are always critical and determinant. However, the efforts to personalize treatment on an individual basis have been quite limited. Currently, there is no intelligent treatment available for this disorder. This research focuses on processing the vital signals of individuals with sleep apnea. Essentially, steps have been taken towards achieving an intelligent treatment. After conducting a comprehensive review of past literature, various vital signals, such as respiratory signals, cardiac activity, and brain activity, were examined. After much effort and communication with researchers to obtain the valuable APPLES dataset, the first step towards intelligent treatment in this thesis is to estimate the apnea severity index. In this section, data from 120 individuals were used, employing three vital signals: heart rate, blood oxygen saturation percentage, and abdominal movements. After pre-processing the signals, relevant features were extracted, and using machine learning algorithms, the subjects were classified into three groups: normal, apnea, and hypopnea. According to global standard criteria, these three events are sufficient to estimate the severity of apnea. The results section shows that we were able to classify patients into the appropriate groups with a 75% accuracy for the three events. After estimating apnea severity, we examined the factors that exacerbate apnea in patients. Initially, we investigated the impact of smoking on the respiratory system of apnea patients. According to the literature, smoking exacerbates this disease, but research in this area is quite limited. By analyzing the signals of 60 individuals (30 smokers and 30 non-smokers) using three vital signals—submental muscle activity, pharyngeal respiratory effort, and intra-nasal mask pressure—we aimed to study the effect of smoking on individuals. In this section, we attempted to differentiate between these individuals during all sleep stages and events using machine learning algorithms. In the corresponding results, we observed significant differences between the two groups during deep sleep in all three signals. The accuracy of the models in various stages ranged from 70% to 95%. Additionally, we compared our results with a paper that attempted to differentiate smokers from non-smokers, finding that our results were superior in the best-case scenario. In the next part of the thesis, we similarly examined the impact of alcohol consumption. Data from 30 individuals (16 alcoholics and 14 non-alcoholics) were analyzed. Based on the previous section's results, we focused on differentiating these two groups during deep sleep and wakefulness. In this part, we used cardiac activity signals, first extracting well-known cardiac signal features and applying SVM and Random Forest models. The best results were around 69%. To improve the results, we used deep learning with a one-dimensional convolutional network, significantly enhancing the results. During deep sleep, we achieved an accuracy of 88%. In the final part of the thesis, we aimed to identify brain connectivity during apnea and wakefulness events using two methods: Granger causality and mutual information. In the APPLES dataset, brain activities were recorded using electrodes C3, C4, O1, and O2. In this section, we first determined brain connectivity over time using the Granger causality metric. To differentiate brain connectivity in two states, we fed the Granger causality values over time as a series input to a one-dimensional convolutional model, achieving an average accuracy of 75%. Then, using the nonlinear method of mutual information, we constructed connectivity matrices. Subsequently, we differentiated the connectivity matrices in the two states (apnea and wakefulness) using a two-dimensional convolutional network, achieving an average accuracy of 85%.
connectivity matrices. Subsequently, we differentiated the connectivity matrices in the two
states (apnea and wakefulness) using a two-dimensional convolutional network, achieving
an average accuracy of 85% - Keywords:
- Signal Processing ; Artificial Intelligence ; Deep Learning ; Machine Learning ; Sleep Apnea ; Sleep Disorder Breathing (SDB)
-
محتواي کتاب
- view