Improving topological performance of automated airway segmentation for longitudinal assessment of the airway tree

• Grant number: 24/22948-0
• Support Opportunities: Scholarships in Brazil - Master
• Start date: May 01, 2025
• End date: July 31, 2026
• Field of knowledge: Engineering - Electrical Engineering
• Principal Investigator: Leticia Rittner
• Grantee: Arthur Matheus do Nascimento
• Host Institution: Faculdade de Engenharia Elétrica e de Computação (FEEC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Company: Universidade Estadual de Campinas (UNICAMP). Faculdade de Engenharia Elétrica e de Computação (FEEC)
• Associated research grant: 20/09838-0 - BI0S - Brazilian Institute of Data Science, AP.PCPE

Abstract

Airway segmentation of CT images has become an important diagnostic tool. By providing a 3D map of the region, it helps us understand how diseases or treatments are affecting its structure. It also plays an important role in surgical planning, facilitating more precise and efficient procedures. However, manual segmentation is a time-consuming and labor-intensive task. The advent of deep learning architectures capable of automatically generating highly accurate segmentations has changed this, enabling the development of methods that significantly streamline and improve the efficiency of CT analysis, but airway segmentation is a complex task that extends beyond simple 2D analysis. Evaluating accuracy slice-by-slice is insufficient, as maintaining volumetric coherence is crucial. While overlapping metrics like DICE are valuable, they alone do not fully capture the quality of a segmentation model, with recent literature exposing the gap in topological accuracy of deep learning based methods. This project aims to advance the state-of-the-art in automated airway segmentation by exploring the implementation of transformer models. By leveraging transformers, we seek to develop a more topologically accurate method for airway segmentation, while maintaining the high overlap accuracy achieved by current methods. Additionally, as part of a BEPE proposal, we will investigate the potential of applying the developed architecture to a longitudinal automated analysis, attempting to provide insights into airway structural changes observed in CT scans across different cancer radiotherapy treatment stages.