تشخیص بیماری آلزایمر با شبکه عصبی کانولوشنی از تصویربرداری تشدید مغناطیسی (مقاله علمی وزارت علوم)
درجه علمی: نشریه علمی (وزارت علوم)
آرشیو
چکیده
مقدمه: این مقاله روشی جدید برای تشخیص بیماری آلزایمر بر اساس ویژگی های تصاویر تصویربرداری تشدید مغناطیسی (MRI) ارائه می کند. تصاویر MRI با حداقل 3 تسلا و ضخامت 3 میلی متر برای تعیین پلاک های پیری و کلاف های مارپیچی ثبت می شود. روش کار: ویژگی های تصاویر MRI مانند آتروفی لوب گیجگاهی میانی، حجم ماده سفید، حجم ماده خاکستری، مایع مغزی نخاعی و عدم تقارن تعیین می شود. افراد به سه گروه افراد سالم، بیماران خفیف و شدید تقسیم شدند. عدم تقارن و میانگین آتروفی لوب گیجگاهی با پیشرفت بیماری آلزایمر افزایش می یابد، زیرا میزان آسیب به لوب گیجگاهی در تصاویر MRI بیماری آلزایمر افزایش یافته است. یافته ها: صحت نتایج شبکه عصبی Elman با ویژگی های استخراج شده از تصاویر MRI با صحت نتایج شبکه عصبی کانولوشنی مقایسه می شود. صحت نتایج با ترکیب ویژگی ها در افراد سالم 82/5 درصد بود. در بیماران آلزایمر خفیف 86/5 درصد و در بیماران آلزایمر شدید 94/5 درصد است. نتیجه گیری: بالاترین صحت نتاج در گروه بیماران مبتلا به آلزایمر شدید و مناسب ترین ویژگی در بین ویژگی های تصاویر MRI، میزان آتروفی لوب گیجگاهی داخلی است. استفاده از شبکه عصبی کانولوشنی نشان می دهد که صحت نتایج در گروه سالم 98 درصد، در گروه خفیف 97/7 درصد و در گروه بیماران شدید 97/5 درصد است. این نتایج نشان می دهد که عملکرد شبکه عصبی کانولوشنی در مقایسه با شبکه عصبی Elman دارای صحت نتابج بیشتری است.Diagnosis of Alzheimer’s disease with convolutional neural network from magnetic resonance imaging
Introduction
Alzheimer’s disease is a progressive disease of the mental powers commonly seen in the elderly. Significant symptoms of this disease are memory loss, judgment, and essential behavioral changes in the person. The disease results in the loss of synapses of neurons in some areas of the brain, the necrosis of brain cells in different areas of the nervous system, the formation of spherical protein structures called aging plaques outside neurons in some areas of the brain, and fibrous protein structures called coils. The helix is identified in the cell body of neurons )1(. The prevalence of Alzheimer’s disease is on the rise. The expenses associated with treating, caring for, and nursing individuals with this condition are substantial and challenging.)2(. This study’s main purpose and motivation was to design and present a method for diagnosing Alzheimer’s disease. Medical image processing plays a crucial role in identifying severe Alzheimer’s disease, as it enables the determination of disease presence and the assessment of similarity between medical images. Additionally, it allows for examining the rate at which the disease progresses.
If this disease is not identified in time, new and up-to-date treatment methods will not work. The solution is to accurately identify the mechanism of this disease and its effect on medical images, which is a challengingtask due to the dynamic nature of the brain and the complex nature of this disease.
Methods
This study recruited fortyvolunteers to record medical images in healthy, mild, and severe groups. The number of people in the healthy group is 19, in the mild group is 11, and in the severe group is ten. Furthermore, 128 image slides were prepared for each candidate. In preparing the images, the MRI machine should have at least 3 Tesla, and the thickness of the slices is 3 mm. The number of slices is 128 in order to see acceptable images to examine the lesions of aging coils and spiral plaques )11(. The database of the Tehran Imaging Center has been prepared. The appropriate image segmentation, mask filter to remove noise and sharp filter to detect the edge in the image was used) 12(. All the data was fed to an augmentation layer before feeding into the neural network. Only the zooming and flipping were selected for augmentation.
Although CT scans are still used regularly for diagnostic evaluation and to study the relationship between the brain and the subject's behavior, they are mostly used when MRI is prohibited because MRI is now the method of choice for evaluating patients. In patients diagnosed with Alzheimer’s disease, it is evident that the inner part of the temporal loop undergoes atrophy, which is associated with nerve lesions.)13(. In most Alzheimer’s patients, atrophy of the inner part of the temporal lobe is detected in patients compared with healthy individuals up to several years before the onset of clinical signs of cognitive impairment. In patients with Alzheimer’s, hippocampal atrophy occurswith a reduction of 10 to 50% and para hippocampus with up to 40% compared to healthy individuals. Clinically, a reduction in hippocampal volume of up to 25% causes mild Alzheimer’s disease. In Alzheimer’s disease, atrophy of the internal temporal lobe and parietal lobe is seen (14). Due to the large number of conditions that can lead to the degeneration of higher structures of the brain, atrophy of the cerebral cortex is one of the most studied types. These causes include a wide range of neurodegenerative diseases, such as Alzheimer’s disease, whose primaryeffect is the destruction of nerve cells and the consequent loss of brain mass (15).
Results
The Elman neural network is used by features extracted from MRI images such as cerebrospinal fluid, white and gray matter volume, atrophy, and asymmetry. As the network architecturewas not complex, the input data after augmentation was enough for training the network. The results were the most accurate when these combined features were used as inputs for the neural network. The actualpositive rate of the results by spinal cord fluid in healthy individuals was 79.9% in mild Alzheimer’s patients, 83.2%, and in severe Alzheimer’s patients, 91%. The true positive rate of the results by the volume of gray matter in healthy individuals was 81.8% in mild Alzheimer’s patients, 84.9%, and in severe Alzheimer’s patients, 92%. The actualpositive rate of the results by white matter volume characteristic in healthy individuals was 82% in mild Alzheimer’s patients 86.2% and in severe Alzheimer’s patients 93%. The results’ actual positive rate by the temporal lobe atrophy characteristic in healthy individuals was 83.3% in mild Alzheimer’s patients, 87.8%, and in severe Alzheimer’s patients, 94.1%. The valid positive rate of the results by asymmetry characteristic in healthy individuals was 80.9% in mild Alzheimer’s patients, 85%, and in severe Alzheimer’s patients, 89.6%. Finally, the validpositive rate of the results by combining the features in healthy individuals was 82.5% in mild Alzheimer’s patients, 86.5%, and in severe Alzheimer’s patients, 94.5%. The highest accuracy of progeny in the group of severe Alzheimer’s patients and the most appropriate feature among the features of MRI images was the degree of medial temporal lobe atrophy. The highest accuracy of prognostication in the group of patients with severe Alzheimer’s and the most suitable feature among the features of MRI images was the amount of atrophy of the medial temporal lobe. Figure 2 shows the accuracy of the neural network results for three groups.
Figure 2. Accuracy of Elman neural network results by MRI image feature
Conclusion
Initially, mask and sharp filters were used to extract the high and low-frequencycomponents of noise in order to extract the appropriate feature and accuracy of the classifier performance. The thickness of the slices in this study in MRI images is considered 4 mm, and the thickness between 0.8 and 4 mm is used to diagnose mild Alzheimer’s disease. On the other hand, the MRI device in this study was 3 Tesla. At least 1.5 to 3 Tesla devices should be used to observe the lesions of aging coils and spiral plaques and to diagnose Alzheimer’s disease. Based on choroidal dislocation, temporal width, and hippocampal height in MRI images, the mean of temporal lobe atrophy in individuals is classified from zero, meaning a healthy condition, to grade 4 with severe cognitive deficits. Using MRI images to diagnose severe Alzheimer’s disease can be an effective way. The selection of appropriate features in these images, such as temporomandibular atrophy, white matter volume, gray matter volume, cerebrospinal fluid, and asymmetry, is appropriate to distinguish healthy individuals from mild and severe. By combining the appropriate features, the precision of neural network outcomes is enhanced. As Alzheimer’s disease advances to severe stages, these features become more distinguishable in comparison to the healthy group and individuals with mild symptoms.This is because, in the initial phases of Alzheimer’s disease, there are minimal atrophy and asymmetry alterations, with no significant variations in white and gray matter volume, as well ascerebrospinal fluid.
Ethical Considerations
Compliance with ethical guidelines
All the subjects received information about the research. They were assured that all information would remain confidential and only be used for research purposes. Furthermore, the patients were assured that if they did not want to participate in the research, it would not lead to the termination of medical services in that center. In order to respect privacy, the details of the subjects were not recorded. Informed consent was obtained from the participants prior to conducting the research, and the researchers observed ethical considerations during the study. Notably, this research has an ethics code with the IR.IMAMREZA.REC.1402.009 at the Imam Reza International University.
Authors’ contributions
All authors contributed to the study’s conception and design. Elias Mazrooei Rad, Sayyed Majid Mazinani, and Hadi Akbari performed material preparation, data collection, and analysis. Elias Mazrooei Rad wrote the first draft of the manuscript, and all authors commented on previous versions. All authors read and approved the final manuscript.
Funding
No financial assistance has been received from any organization.
Acknowledgments
The authors express their gratitude to Kavian Ghandehari for her cooperation with the patients in registering the signal.
Conflict of interest
The authors declared no conflict of interest.
تبلیغات
