Zakia Khatun will present her Ph.D. dissertation on the use of Artificial Intelligence (AI) in multi-modal medical imaging to improve the assessment of soft tissue pathologies, specifically in tendon-related conditions. This research focuses on advanced AI techniques, such as deep learning, to automate the analysis of medical images from modalities like MRI and CT. By implementing innovative methods for tendon segmentation, pathology classification, and reflex response assessment, the study aims to enhance diagnostic accuracy and support clinical decision-making. Khatun's work represents a significant step towards more efficient and reliable healthcare interventions, with potential applications in managing musculoskeletal health.

• Student: Zakia Khatun
• Supervisor: Professor Paolo Gargiulo, Reykjavík University
• Co-Supervisor: Professor Halldór Jónsson jr, Emeritus, Department of Engineering, Institute for Biomedical and Neural Engineering

Committee

• Professor Francesco Tortorella, University of Salerno, Department of Information and Electrical Engineering and Applied Mathematics 
• Associated Professor Þórður Helgasson, Reykjavik University, Department of Engineering

Examiner

• Prof Cristiana Corsi, University of Bologna

Teams:

• Join the meeting now
• Meeting ID: 315 230 483 618
• Passcode: mQPiZK

Abstract
Artificial Intelligence (AI) has transformed many fields by automating complex tasks, identifying patterns in large datasets, and making accurate predictions. Machine learning, particularly deep learning, enables AI to mimic human intelligence for tasks such as image segmentation, classification, and recognition. In medical imaging, modalities like Magnetic Resonance Imaging (MRI), Computed Tomography (CT), and Ultrasound are crucial for diagnosing pathologies, detecting abnormalities, guiding treatment, and monitoring disease progression.

Integrating AI with medical imaging can improve the assessment and management of tendon-related conditions, such as tendinopathy, which can significantly impact quality of life. Early detection is vital to optimize treatment outcomes, and this thesis aims to develop advanced methods for analyzing soft tissue pathologies, particularly tendon segmentation, pathology classification, and tendon reflex response assessment. By automating the analysis of soft tissues, these methods seek to reduce human error and variability, supporting more reliable clinical decisions and earlier interventions.

The first study analyzes MRI and CT images of 47 participants to explore the relationships between tendons, cartilage, and muscles. This study shows that knee cartilage degeneration can be predicted even using features from the quadriceps and patellar tendon only. Traditional machine learning models are used to identify key features associated with both tendinopathy and cartilage degeneration predictions. This foundational work improves understanding of soft tissue relationships, contributing to better diagnostic approaches in musculoskeletal health.

Another key component of this thesis is the development of an end-to-end tendon segmentation module. This system includes a superpixel-based coarse segmentation step to aid final segmentation. Two approaches are compared: (1) Random Forest (RF) and Support Vector Machine (SVM) classifiers for superpixel categorization, and (2) a Graph Convolutional Network (GCN) for transforming superpixels into graph structures for classification. The RF and SVM methods achieve Area Under the Curve (AUC) scores of 0.992 and 0.987, respectively, with high sensitivity. The GCN approach, while slightly less effective, demonstrates the potential of superpixel generation in improving segmentation.

Another study focuses on developing an end-to-end tendon pathology detection module, using the same MRI dataset. A graph-based model, where superpixels act as nodes, is implemented using a Graph Echo State Network (GESN) for classification. The GESN, which efficiently represents data without iterative backpropagation, outperforms traditional machine learning models, achieving a mean accuracy of 0.953 and sensitivity of 0.943, highlighting its potential for improving diagnostic accuracy.

The last study investigates how age, height, weight, and gender affect reflex response times in healthy individuals using electromyography (EMG) recordings from 40 participants. Results show that elderly individuals, particularly taller, heavier, and male participants, have delayed reflex onsets. Even with height normalization, elderly participants show slower reflexes. Younger participants have longer total reflex durations, likely due to their height, a pattern consistent across genders. These findings emphasize the influence of demographic factors on neuromuscular reflexes and may help in diagnosing neuromuscular disorders.

In conclusion, this research demonstrates the potential of AI, particularly superpixel-based and graph-based models, to advance tendon pathology assessment and exploratory tendon reflex studies, leading to better patient outcomes and musculoskeletal health management.