University of Thessaly (UTH)
Αξιοποίηση της βαθιάς μάθησης για την ακριβή διάγνωση της ρήξης του πρόσθιου χιαστού συνδέσμου και την ανίχνευση ανωμαλιών της άρθρωσης του γονάτου στην ιατρική απεικόνιση
Abstract
dc:descriptionThe accurate diagnosis of anterior cruciate ligament (ACL) rupture is crucial for the effective treatment and rehabilitation of knee injuries, which are particularly prevalent among athletes. Recent advancements in deep learning (DL) have significantly improved the detection of ACL injuries using MRI studies. This dissertation aims to explore and enhance the application of DL techniques for the precise identification and evaluation of ACL injuries in medical imaging. A systematic review was conducted to summarize existing DL methodologies for detecting knee injuries, highlighting the strengths and limitations of current approaches. Building upon this foundation, a lightweight DL pipeline using YOLOv5-Nano was developed to identify healthy ACLs in sagittal MRI slices, achieving high accuracy and efficiency with minimal computational resources. Additionally, the thesis introduces the Thessaly Graft Index (TGI), an AI-based index for assessing graft healing post-ACL reconstruction. The TGI demonstrated a strong correlation with clinical and functional outcomes, proving its potential as a reliable tool for postoperative evaluation. To address class imbalance and enhance diagnostic precision, an innovative hybrid DL framework was formulated, reconfiguring ACL tear classification as a novelty detection problem. This approach leveraged aleatoric semantic uncertainty and demonstrated superior performance across multiple datasets, including cross-database testing with public and clinical MRI data. The dissertation's findings underscore the potential of DL to match human-level performance in MRI-based knee injury diagnosis. Key contributions include the development of resource-efficient models, the introduction of novel diagnostic indices, and the demonstration of robust generalization capabilities across diverse datasets. These advancements pave the way for the widespread clinical application of DL in the accurate and cost-effective diagnosis of ACL injuries, promising significant improvements in patient outcomes and healthcare efficiency.
Degree
thesis:*- Grantor dc:publisher
- University of Thessaly (UTH)
- Year dc:date
- 2024
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Siouras, Athanasios
- Σιούρας, Αθανάσιος
Subjects
dc:subject × 25- Τεχνητή νοημοσύνη (ΤΝ)
- Επεξεργασία εικόνας
- Ορθοπεδική
- Μαγνητική τομογραφία γόνατος
- Ανίχνευση αντικειμένων
- Βαθιά μάθηση
- Object detection and recognition
- Anterior cruciate ligament
- Computer vision and image processing
- Deep learning
- Diagnostic classification methods
- Knee
- Orthopedics
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Τεχνητή νοημοσύνη
- Κλινική Ιατρική
- Ιατρική και Επιστήμες Υγείας
- Ορθοπεδική και Αθλιατρική
- Computer and Information Sciences
- Natural Sciences
- Artificial Intelligence
- Clinical Medicine
- Medical and Health Sciences
- Orthopedics and Sports Medicine
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/58219