Luisa Vargas Daza
Segmentation of treelike tubular structures in medical imag- ing, crucial for accurate diagnosis and treatment. Tradi- tional methods often struggle with the complex morphology and inherent data variability of structures like blood vessels and lung branching. To tackle these challenges, this work presents three significant contributions. First, it introduces a comprehensive dataset aggregation, focusing on tubular structures, to challenge and benchmark existing segmenta- tion algorithms. Second, an innovative evaluation frame- work is developed, surpassing traditional metrics by ac- curately assessing segmentation quality based on geomet- rical and topological characteristics of tubular structures. Lastly, the thesis proposes the Joint Brain-Vessel Segmen- tation (JoB-VS) framework, an end-to-end solution for seg- menting brain vessels in TOF-MRA images, enhancing per- formance by forgoing additional preprocessing steps. These contributions collectively advance the field of medical im- age analysis, bridging the gap between technical segmen- tation techniques and their clinical application, thereby en- hancing diagnostics and treatment planning in healthcare.