Aristotle University Of Thessaloniki (AUTH)
Ανάλυση δεδομένων με τεχνικές μηχανικής και βαθιάς μάθησης για τη μοντελοποίηση και εκτίμηση της εξέλιξης της οστεοαρθρίτιδας γονάτου σε ασθενείς
Abstract
dc:descriptionKnee osteoarthritis is one of the most common musculoskeletal disorders, affecting millions of people worldwide. The disease is characterized by progressive degeneration of the articular cartilage, accompanied by pain, joint stiffness, and limited joint mobility. Early diagnosis and monitoring of the progression of osteoarthritis are crucial for its effective management and improving patients’ quality of life. Medical imaging such as X‐rays and magnetic resonance imaging (MRI) are some of the most reliable tools available to healthcare professionals for imaging soft tissues, including articular cartilage, making them ideal for assessing osteoarthritis and evaluating the progressive course of the disease. The purpose of this thesis is twofold: On the one hand, the first objective is to develop machine and deep learning methods for the automatic segmentation of knee articular cartilage (femoral and tibial) from magnetic resonance images. On the other hand, using the techniques developed in the first part to produce accurate segmentations, the goal is to develop longitudinal models for assessing the risk of osteoarthritis, as well as forecasting the course of the disease once it has already developed. This thesis synthesizes six published research papers organized into three parts that follow a unified methodological trajectory: (I) cartilage segmentation via multi‐atlas registration and semi‐supervised learning; (II) cartilage segmentation via graph and hypergraph convolutional networks; and (III) longitudinal prediction of disease progression via spatio‐temporal hypergraph networks. In the first stage, a family of algorithms is proposed for the automatic segmentation of magnetic resonance imaging using a patch‐based label propagation method. Subsequently, the above process is integrated into the framework of deep learning through the use of innovative architectures based on graph and hypergraph neural networks. Finally, the above models are enriched to enable the assessment of disease progression through the introduction of the temporal component into their architectural framework. The output of the segmentation stages feeds directly into the prediction stages, via a three‐dimensional representation of cartilage as a volumetric‐node graph from which a compact shape descriptor is extracted. The proposed methods are applied to an extensive data repository (Osteoarthritis Initiative — OAI) which, among other things, includes magnetic and radiological images from patients who have been monitored over a number of years, as well as additional demographic data (age, gender, body mass index), which can contribute to a more accurate assessment of the course of the disease. The results obtained from the application of the proposed methods are encouraging: for cartilage segmentation, the methods achieve a Dice similarity coefficient of up to 96%, comparable to the agreement of expert radiologists; for longitudinal progression prediction, they achieve an AUC of approximately 0.94 over prediction horizons of up to 8 years, depending on the available history. These results demonstrate that the combined exploitation of structural information via graphs and hypergraphs, multiple complementary views, and spatio‐temporalmodeling constitutes a promising direction for personalized computational medicine in knee osteoarthritis. Keywords: knee osteoarthritis, cartilage segmentation, magnetic resonance imaging, semi‐supervised learning, graph neural networks, hypergraphs, spatio‐temporal models, longitudinal prediction.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Chadoulos, Christos
- Χαδουλός, Χρήστος
Subjects
dc:subject × 18- Μηχανική μάθηση
- Βαθιά μάθηση
- Ψηφιακή επεξεργασία ιατρικής εικόνας
- Κατάτμηση εικόνων
- Χωροχρονική εκτίμηση
- Νευρωνικά δίκτυα γράφων (GNNs)
- Machine learning model classification and training
- Medical image processing techniques
- Deep learning
- Hypergraph learning
- Graph neural networks (GNNs)
- Spatiotemporal modelling
- Επιστήμη Ηλεκτρολόγου Μηχανικού, Ηλεκτρονικού Μηχανικού, Μηχανικού Η/Υ
- Επιστήμες Μηχανικού και Τεχνολογία
- Αυτοματισμοί και Συστήματα ελέγχου
- Electrical Engineering, Electronic Engineering, Information Engineering
- Engineering and Technology
- Automation and control systems
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/62873