Abstract
بیماری کبد چرب غیر الکلی از شایع ترین بیماری های کبدی است که در صورت تشخیص به موقع، قابل درمان است. تصویر برداری فراصوت یک روش غیرتهاجمی و در دسترس عموم در انر تشخیص و غاربالگری کبد چرب است. اما به علت پیچیدگی فرآیند تشخیص، میزان تخصص و مهارت رادیولوژیست بر تفسیر نتایج اثرگذار است. در این راستا روش های استخراج ویژگی فرکانسی نتایج دقیق تری نسبت به ویژگی های مبتنی بر بافت ارائه داده اند. اما محدودیت این روش ها، عدم یکنواختی عملکرد آن ها در پایگاه های داده مختلف است. در این پژوهش با هدف دسته بندی بیماران کبد چرب و سطح آن از روش استخراج ویژگی مبتنی بر یادگیری عمیق با استفاده از آموزش شبکه ی AlexNet همراه با دسته بندی کننده SVM استفاده شده است. در این پژوهش از دو پایگاه داده استفاده شده است و نتایج برای پایگاه داده ی آوزش داده شده و پایگاه داده ی کاملاً جدید به ترتیب 96% و 90% برای تفکیک بیماران از افراد سالم و 92% و 83% برای تفکیک سطوح چربی به دست آمد. این روش نتایج پایدارتری را روی دو پایگاه داده ی مختلف نسبت به روش های فرکانسی نشان می دهد که به این ترتیب می توان نتیجه گرفت بازگشت به حوزه مکان فضای تصویر با استفاده ویژگی های یادگیری عمیق می تواند نتایج پایدارتری را با دقت مشابه به همراه داشته باشد.
Abstract in English:
Non-alcoholic fatty liver disease is one of the most common liver diseases that can be treated if diagnosed on time. Ultrasound imaging is a non-invasive method available to the public in diagnosing and diagnosing fatty liver. But due to the complexity of the diagnosis process, the level of expertise and skill of the radiologist affects the interpretation of the results. In this regard, frequency feature extraction methods have provided more accurate results than texture-based features. But the limitation of these methods is the non-uniformity of their performance in different databases. In this research, with the aim of classifying patients with fatty liver and its level, feature extraction method based on deep learning was used using AlexNet network training along with SVM classifier. In this research, two databases have been used, and the results for the updated database and the completely new database are 96% and 90% respectively for separating patients from healthy people and 92% and 83% for separating fat levels in the hand. came This method shows more stable results on two different databases than frequency methods, so it can be concluded that returning to the area of image space using deep learning features can bring more stable results with similar accuracy.
Abstract in English:
Non-alcoholic fatty liver disease is one of the most common liver diseases that can be treated if diagnosed on time. Ultrasound imaging is a non-invasive method available to the public in diagnosing and diagnosing fatty liver. But due to the complexity of the diagnosis process, the level of expertise and skill of the radiologist affects the interpretation of the results. In this regard, frequency feature extraction methods have provided more accurate results than texture-based features. But the limitation of these methods is the non-uniformity of their performance in different databases. In this research, with the aim of classifying patients with fatty liver and its level, feature extraction method based on deep learning was used using AlexNet network training along with SVM classifier. In this research, two databases have been used, and the results for the updated database and the completely new database are 96% and 90% respectively for separating patients from healthy people and 92% and 83% for separating fat levels in the hand. came This method shows more stable results on two different databases than frequency methods, so it can be concluded that returning to the area of image space using deep learning features can bring more stable results with similar accuracy.
Translated title of the contribution | Diagnosis of fatty liver disease based on feature extraction from AlexNet network trained on liver ultrasound images |
---|---|
Original language | Persian (Iran, Islamic Republic of) |
Title of host publication | ششمین کنفرانس پردازش سیگنال و سیستمهای هوشمند |
Place of Publication | Iran |
Publisher | Civilica |
Number of pages | 7 |
Publication status | Published - 2020 |
Event | 6th Conference on Signal Processing and Intelligent Systems - Khorasan Razavi, Mashhad, Iran, Islamic Republic of Duration: 22 Dec 2020 → 22 Dec 2020 https://en.civilica.com/l/11563/ |
Conference
Conference | 6th Conference on Signal Processing and Intelligent Systems |
---|---|
Country/Territory | Iran, Islamic Republic of |
City | Mashhad |
Period | 22/12/20 → 22/12/20 |
Internet address |
Keywords
- non-alcoholic fatty liver disease
- ultrasound image processing
- fatty liver screening
- deep learning in convolution network ultrasound image feature extraction