Please use this identifier to cite or link to this item: http://hdl.handle.net/10609/139427
Title: Automatic detection of knee joints and classification of knee osteoarthritis severity from plain radiographs using CNNs
Author: Durán Olivar, David
Tutor: Espinós Morato, Héctor
Others: Casas-Roma, Jordi  
Abstract: Knee Osteoarthritis (OA) is the most common type of arthritis and it is typically the result of wear and tear, and progressive loss of articular cartilage which may eventually lead to disability. OA diagnosis is typically conducted by performing a physical examination of the knee by means of a visual inspection of radiographic imaging. Based on the presence of OA pathological features of joint space narrowing, osteophyte formation or sclerosis, Kellgren-Lawrence (KL) system is typically used to classify the severity of the disease into one of five ranked grades. The conclusion regarding the presence and severity of knee OA may differ due to the subjective nature of the assessment. In this study, we present a computer-aided diagnosis method based on Convolutional Neural Networks (CNN) to automatically locate and score knee OA severity from X-ray images according to the KL grading scale. Location of the knee joints is achieved by considering a region of interest (ROI) segmentation with U-Net architecture. Transfer learning from pre-trained CNN architectures is considered for knee OA severity assessment. Our method yields a quadratic Cohen Kappa coefficient of 0.87 and a weighted average f1-score of 72%. In addition, we show attention maps highlighting the strongest contribution to the network prediction. The visualization provides practitioners with information to assist in the diagnosis decision-making processes. We conclude that our methodology achieves high accuracy in localizing knee joints out of plain X-ray images, as well as a very good performance in the OA diagnostic assessment of KL grades.
Keywords: knee osteoarthritis
U-Net
convolutional Neural Networks
Document type: info:eu-repo/semantics/masterThesis
Issue Date: Jan-2022
Publication license: http://creativecommons.org/licenses/by-nc-nd/3.0/es  
Appears in Collections:Bachelor thesis, research projects, etc.

Files in This Item:
File Description SizeFormat 
davidduranTFM0122memory.pdfTFM memory9,93 MBAdobe PDFThumbnail
View/Open