Application of artificial intelligence for the lumen segmentation in Οct images
Loading...
Date
2025-10-20
Authors
Κεκάκου, Μαργαρίτα
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Lumen segmentation is fundamental for the diagnosis, monitoring and treatment of coronary artery disease. Optical coherence tomography (OCT) provides high-resolution intravascular images, but OCT lumen segmentation is typically performed manually, requiring time and expertise due to lumen morphology and OCT-specific artifacts. This thesis develops and evaluates an artificial intelligence (AI) framework for automated lumen segmentation in intravascular OCT images. The proposed model is a lightweight convolutional neural network (CNN) with an encoder-decoder architecture, designed to balance accuracy and computational efficiency. During the development phase, different design choices are systematically examined, including target representation, optimizers, and data augmentation strategies, in order to identify the most effective configuration. The final framework integrates lumen and adventitia prediction, the AdamW optimizer, and a tailored augmentation pipeline. Evaluation is performed using patient-level six-fold cross-validation to prevent data leakage and provide reliable estimates of generalization, with segmentation quality assessed through standard evaluation metrics and visualization of predictions. A resolution study is also conducted to investigate the impact of input size on performance. Finally, considerations of training and inference time are included to highlight the trade-off between segmentation accuracy and computational cost. Overall, the study demonstrates that accurate and efficient lumen segmentation is achievable with a lightweight CNN and a limited dataset through careful design choices, highlighting the clinical potential of AI methods in real-time OCT analysis.
Description
Keywords
Coronary artery disease, Optical coherence tomography, Lumen segmentation, Convolutional neural networks, Artificial intelligence