University of Oulu

The role of bone marrow lesions in knee osteoarthritis : textural analysis of subchondral bone

Saved in:
Author: Kazemtarghi, Amir1
Organizations: 1University of Oulu, Faculty of Information Technology and Electrical Engineering, Computer Science
Format: ebook
Version: published version
Access: open
Online Access: PDF Full Text (PDF, 10 MB)
Pages: 59
Persistent link:
Language: English
Published: Oulu : A. Kazemtarghi, 2020
Publish Date: 2020-08-18
Thesis type: Master's thesis (tech)
Tutor: Yalcin Bayramoglu, Neslihan
Panfilov, Egor
Reviewer: Saarakkala, Simo
Yalcin Bayramoglu, Neslihan


Osteoarthritis (OA) is the most common joint disorder in the world that affects various joints particularly hand, hip, and knee joint. The knee OA has been identified as the most impactful OA because it is the major cause of disability worldwide. Generally, OA progression leads to joint replacement surgery and causes enormous amount of financial costs. Thus, it is crucial to diagnose OA at an early stage and prevent or slow down its progression. Currently, clinical diagnosis of OA includes physical examination and clinical imaging. However, they are insensitive to early OA changes. On the other hand, it was shown that several imaging bio markers can be captured at an early stage of the disease. One of the important imaging bio markers for OA is the alternations of subchondral bone texture. Besides, there are other factors that cause these alternations such as bone marrow lesions (BML).

Two sub-studies have been conducted in this thesis. The aim of the first sub-study is to investigate the association between BML and OA diagnosis by using subchondral bone texture from plain radiography. OA subjects are defined by Kellgren-Lawrence (KL) grading scale. KL grade 0 and 1 represent no OA and grade 2, 3, and 4 are OA subjects. In this work, subjects at the baseline (first visit) of osteoarthritis initiative (OAI) dataset were selected. Then, they were categorised into three groups including subjects who has BML in medial tibia (group 1), subjects without BMLs at all (group 2), and lastly the subjects without medial tibia BMLs (group 3). In the next step, region of interest (ROI) was selected at the margin of medial tibia in plain radiographs. After that, 29 textural features from 4 textural descriptors including grey-level co-occurrence matrix (GLCM), histogram of image, absolute gradient, and fractal signature analysis (FSA) were computed from the extracted ROI. Subsequently, Fisher’s exact test and Mann–Whitney U test were used in order to discover how textural features change among OA and non OA subjects in each group (first analysis) and how those differences change across the groups (second analysis).

Our results showed that there are significant textural differences between OA and non OA subjects when they have BMLs at medial tibia. Moreover, there were no significant textural differences among subjects with no BMLs and subjects with no BMLs in medial tibia. These results indicate that the presence of BML as well as its location at subchondral bone may have association with OA incidence.

In the second sub-study, for research oriented purposes we built a deep convolutional neural network (CNN) based models to automatically detect OA from subchondral bone texture according to the Kellgren-Lawrence (KL) grading scale. We selected subjects without BMLs to make a fair comparison between magnetic resonance imaging (MRI) data and plain radiographs. In this study, subjects with no BMLs who have sagittal 3-D Double Echo Steady State sequence (3D DESS MRI) and plain radiography at the baseline of OAI were selected. In both imaging modalities, square sized ROIs were chosen located at the marginal region of medial tibia. Confusion matrix and area under the receiver operating characteristics curve (ROC AUC) were used to evaluate the model performance. Our results demonstrated that when subjects do not have BMLs, our model was not able to detect OA from the subchondral bone texture.

see all

Copyright information: © Amir Kazemtarghi, 2020. This publication is copyrighted. You may download, display and print it for your own personal use. Commercial use is prohibited.