P19 Structure-from-motion analysis may generate an accurate automated bowel preparation score

Introduction Structure-from-Motion (SfM) is a computer vision technique which allows us to estimate the 3D structure of a scene from a set of 2D images. Our aim was to use this to automatically identify quality of bowel preparation. Methods We applied SfM to 5 colonoscopy sequences, composed of 150 to 300 consecutive images displaying caecum. We then refined the estimated 3D meshes by smoothing them and eliminating erroneous estimates arising at the edge of the reconstructed surfaces. These erroneous estimates were mainly due to a lack of visual redundancy, motion blur or illumination artefacts such as large specularities. Results Figure 1 shows that SfM allows successful estimation of 3D structure of different caecum sections. Depressed and protruded areas could particularly facilitate visual analysis. Although SfM suffers from a scale ambiguity which prevents 3D measurements, it can provide different quality indicators such as an estimate of the percentage of colonic surface observed during a procedure. Here, we evaluated effectiveness of pre-operative bowel preparation by measuring the ratio of obscured or partially obscured area over the 3D surface reconstructed. Figure 1(b) and figure 1(c) correspond to clean bowel preparation with a percentage of obscured mucosa less than 2%. Figure 1(a) illustrates poor bowel preparation as approximately 20% of the observed colon section is obscured. For some images of the corresponding colonoscopy sequence, 35% of colon surface observed was obscured due to poor bowel preparation. Such a quality indicator would contribute to an objective assessment of colonoscopy examination reliability. Conclusion This study demonstrates that 3D vision-based approaches can provide objective quality indicators in colonoscopy. More advanced approaches such as Simultaneous Localisation And Mapping (SLAM) could also be used to estimate both the 3D structure of the observed scene and the endoscope motion. SLAM could provide practitioners with enhanced visualisation in colonoscopy contributing to the development of advanced quality indicators. Acknowledgement The EndoMapper project has received funding from the European Union’s Horizon 2020 research and innovation program under grant agreement No 86314