Evaluación de técnicas de navegación para colonoscopia virtual

Abstract

Virtual colonoscopy (CTC) is a non-invasive technique that enables the analysis of the colon’s interior using computed tomography images, allowing for the reconstruction of its three-dimensional structure and virtual navigation through the generated 3D model. This approach facilitates internal exploration without invasive procedures, supporting the evaluation of the colonic lumen structure and the development of endoluminal navigation strategies. The purpose of this work is to compare two approaches applied to navigation within three-dimensional colon environments: interactive coverage-guided navigation (ColNav) and automatic navigation based on singularity-free vector fields (SF-GVF). Using a clinical study extracted from the TCIA repository (ACRIN 6664), a 3D model of the colon was reconstructed following a methodological workflow that included lumen segmentation through thresholding and morphological operations in 3D Slicer, surface reconstruction, and centerline extraction with VMTK, which was smoothed using B-spline interpolation to obtain a continuous trajectory of the internal anatomy. On this anatomical model, two navigation techniques were implemented and evaluated. In the case of ColNav, an interactive environment was developed with a two-dimensional unfolded map that facilitated angular visualization of the lumen and motion mapping according to unobserved regions, enabling control over the coverage of the traversal. In contrast, the SFGVF method generated a continuous, singularity-free vector field that guided movement through the lumen while automatically maintaining the direction of advancement. Both approaches were tested under the same experimental conditions to ensure comparability of results, using identical geometry, simulation parameters, and visualization systems. Quantitative evaluation was carried out using metrics reported in recent literature on virtual navigation, allowing the performance of the implemented techniques to be compared in terms of global coverage, deviation from the centerline, redundancy, and total number of steps. The experimental results revealed differences in exploration patterns, with ColNav oriented toward local visibility-dependent decisions, and SF-GVF favoring continuous trajectories with fewer repetitions. This study proposes a reproducible protocol for reconstruction, implementation, and measurement that may serve as a reference for validating new navigation strategies in virtual colonoscopy and for extending these approaches to other algorithms or datasets in three-dimensional colon analysis.