Analysis of Modic Degenerations Detected in Magnetic Resonance Imaging With Deep Learning Techniques
No Thumbnail Available
Date
2022
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Bu çalışmamızda, manyetik rezonans görüntülemede(MRG) saptanan modic dejenerasyon bulgularının derin öğrenme teknikleri kulllanılarak analiz edilmesi amaçlanmıştır. Gereç ve Yöntem: 2016-2021 yıllarında Lomber MRG tetkiki uygulanan, yaşları 19-86 arasında değişen 125'i kadın ve 182'si erkek toplam 307 hastada sagittal T1, sagittal ve aksiyel T2 ağırlıklı lomber MRG görüntüleri incelendi. Modic dejenerasyonlar(MD) sinyal değişikliklerine göre kategorize edilip işaretlendi. Çalışmamız sınıflandırma ve segmentasyon olmak üzere birbirinden bağımsız iki aşamadan oluşmaktadır. Kategorize edilen veriler ilk aşamada DenseNet-121, DenseNet-169, VGG-19 gibi ESA(evrişimli sinir ağı) mimarileri ile sınıflandırılmıştır. Daha sonraki aşamada ise işaretlenen resimler üzerinden resim işleme programlarıyla ESA mimarisi olan U-Net ile segmentasyon yapılarak maskeler çıkarılmıştır. Bulgular: Sınıflandırma aşamasında Modic-1, Modic-2, Modic-3 dejenerasyonlarda başarı oranı sırasıyla DenseNet-121'de %98, %96, %100 , DenseNet-169'da %100, %94, %100 , VGG-19'da %98, %92, %97 bulunmuştur. Segmentasyon aşamasında U-Net mimarisi ile başarı oranı %71 bulunmuştur. Sonuç: Bel ağrısı etiyolojisinde yer alan modic dejenerasyonların MRG bulgularının derin öğrenme mimarileriyle değerledirilmesi teşhis kolaylılığı sağlayarak radyoloji hekiminin iş yükünü önemli ölçüde azaltabilir.
In this study, we aimed to analyze the modic degeneration findings detected in magnetic resonance imaging (MRI) by using deep learning techniques. Material and Method: Sagittal T1, sagittal and axial T2-weighted lumbar MRI images were analyzed in a total of 307 patients, 125 female and 182 male, aged between 19-86 years, who underwent Lumbar MRI examination in 2016-2021. Modic degenerations (MD) were categorized and marked according to signal changes. Our study consists of two independent stages, namely classification and segmentation. The categorized data were first classified with CNN(convolutional neural network) architectures such as DenseNet-121, DenseNet-169, VGG-19. In the next stage, masks were removed by segmentation with U-Net, which is the CNN architecture, with image processing programs on the marked images. Results: At the classification stage, the success rates in Modic-1, Modic-2, Modic-3 degenerations were 98%, 96%, 100% in DenseNet-121, 100%, 94%, 100% in DenseNet-169, in VGG-19 98%, 92%, 97% respectively found. In the segmentation phase, the success rate was 71% with the U-Net architecture. Conclusion: Evaluation of MRI findings of modic degenerations in the etiology of low back pain with deep learning architectures can significantly reduce the workload of the radiologist by providing ease of diagnosis.
In this study, we aimed to analyze the modic degeneration findings detected in magnetic resonance imaging (MRI) by using deep learning techniques. Material and Method: Sagittal T1, sagittal and axial T2-weighted lumbar MRI images were analyzed in a total of 307 patients, 125 female and 182 male, aged between 19-86 years, who underwent Lumbar MRI examination in 2016-2021. Modic degenerations (MD) were categorized and marked according to signal changes. Our study consists of two independent stages, namely classification and segmentation. The categorized data were first classified with CNN(convolutional neural network) architectures such as DenseNet-121, DenseNet-169, VGG-19. In the next stage, masks were removed by segmentation with U-Net, which is the CNN architecture, with image processing programs on the marked images. Results: At the classification stage, the success rates in Modic-1, Modic-2, Modic-3 degenerations were 98%, 96%, 100% in DenseNet-121, 100%, 94%, 100% in DenseNet-169, in VGG-19 98%, 92%, 97% respectively found. In the segmentation phase, the success rate was 71% with the U-Net architecture. Conclusion: Evaluation of MRI findings of modic degenerations in the etiology of low back pain with deep learning architectures can significantly reduce the workload of the radiologist by providing ease of diagnosis.
Description
Keywords
Radyoloji ve Nükleer Tıp, Bel ağrısı, Derin öğrenme, Evrişimli sinir ağları, Makine öğrenmesi, Manyetik alanlar, Manyetik rezonans görüntüleme, Omurga, Sinir ağları, Spinal hastalıklar, Yapay zeka, Radiology and Nuclear Medicine, Back pain, Deep learning, Convolutional neural networks, Machine learning, Magnetic fields, Magnetic resonance imaging, Spine, Nerve net, Spinal diseases, Artificial intelligence
Turkish CoHE Thesis Center URL
WoS Q
Scopus Q
Source
Volume
Issue
Start Page
End Page
46