دانلود مقاله شبکه عصبی عمیق و روش یادگیری ماشین جهت طبقه بندی تصویر فوندوس شبکیه
خلاصه
1. معرفی
2. پس زمینه
3. سیستم پیشنهادی
4. نتایج
5. نتیجه گیری
منابع مالی
اعلامیه منافع رقابتی
قدردانی
در دسترس بودن داده ها
منابع
Abstract
1. Introduction
2. Background
3. Proposed system
4. Results
5. Conclusion
Funding
Declaration of Competing Interest
Acknowledgments
Data availability
References
چکیده
دیابت یک بیماری مزمن شایع و یک مشکل عمده بهداشت عمومی است که به ابعاد اپیدمی در سطح جهانی نزدیک می شود. افراد مبتلا به دیابت بیشتر از افراد بدون دیابت به گلوکوم مبتلا می شوند. اگر گلوکوم در مراحل اولیه تشخیص داده نشود، می تواند منجر به از دست دادن بینایی شود. این مطالعه یک سیستم تریاژ کامپیوتری هوشمند با یک شبکه عصبی عمیق و یادگیری ماشینی را برای توسعه و تجزیه و تحلیل تصاویر رنگی فوندوس شبکیه و طبقه‌بندی تصاویر گلوکوماتوز شبکیه پیشنهاد می‌کند. ویژگی‌های عمیق تصاویر شبکیه از تصویر شبکیه فوندوس با استفاده از یک شبکه عصبی عمیق استخراج می‌شود و طبقه‌بندی ویژگی‌ها با استفاده از طبقه‌بندی‌کننده‌های مختلف یادگیری ماشین انجام و تحلیل می‌شود. نتایج تجربی نشان می‌دهد که ترکیب شبکه عصبی عمیق و طبقه‌بندی‌کننده مبتنی بر رگرسیون لجستیک از همه سیستم‌های تریاژ گلوکوماتوز موجود بهتر عمل می‌کند و دقت طبقه‌بندی، حساسیت و ویژگی را بهبود می‌بخشد.
Abstract
Diabetes is a common chronic disease and a major public health problem approaching epidemic proportions globally. People with diabetes are more likely to suffer from glaucoma than people without diabetes. Glaucoma can lead to loss of vision if not diagnosed at an early stage. This study proposes an intelligent computer-aided triage system with a deep neural network and machine learning to develop and analyze color retinal fundus images and classify glaucomatous retinal images. Deep features of retinal images from the fundus retinal image are extracted using a deep neural network, and the classification of features is performed and analyzed using different machine learning classifiers. Experimental results show that the combination of deep neural network and logistic regression-based classifier outperforms all existing glaucomatous triage systems, improving classification accuracy, sensitivity, and specificity.
Introduction
According to the World Health Organization (WHO) report [1], glaucoma is one of major eye disease which is affecting millions of people in developing countries such as India. The Glaucoma damages the retina in a progressive manner and is less detected by the person and finally causes blindness. As per the survey of glaucoma society of India [2], around 12 million people in India are suffering from this disease. Therefore, early detection and treatment of glaucoma to reduce the risk of blindness is the need of the hour. The acquisition of retinal information of eye is usually performed by gonioscopy and ophthalmoscopy. Then, the analysis of glaucoma for physical condition of the optical nerve is done by filed vision test, intraocular pressure, etc. [3], [4], [5]. The glaucoma detection can be performed using optical coherence tomography (OCT), visual test chart and color fundus camera. The glaucoma detection can also be done by analyzing features such as cup-to-disk (CDR) and disk rim thickness rule of inferior, superior, nasal, and temporal (ISNT).
Recently, computer-aided systems [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18] based on various image processing and machine learning algorithms are gaining importance for intelligent detection of glaucoma [19], [20]. While supervised machine learning algorithms are used for classification of the normal image and glaucomatous image for a given dataset of the retinal image, the unsupervised machine learning algorithms are used mainly for segmentation of disk and cup in the enhanced retinal image. Recently, the various schemes are proposed by various researchers for detection of glaucoma in retinal images.
Conclusion
In this paper, a deep neural network and machine learning based computer-aided system for classification of the glaucomatous retinal image is proposed. Here, 512 deep features of retinal images are explored using DNN. The proposed system tested and analyzed all images from the public datasets such as DRISTHI-GS1 and ORIGA. The classification of the glaucomatous retinal image is performed using six machine learning-based classifiers: kNN, SVM, DT, RF, NB, and LR. It is observed that the combination of DNN with LR based machine learning-based classifier outperforms all existing glaucomatous screening systems, with improvement in classification accuracy and sensitivity.
Future work will be focused on the assessment of deep learning-based classifier combination for classification of glaucomatous retinal images. The work may also focus on detection and screening of vascular bleeding in the retina due to macular degeneration and diabetic retinopathy (DR) using machine learning techniques. Also, experimental results particularly F1 score indicates that proposed system need to test with more balance dataset to improve this score and acceptance for practical implementation in future.