Optimising performance in clinical capsule endoscopy

Abstract: Video capsule endoscopy (VCE), perfomed by ingesting a small vitamin-sized camera pill, was developed over a period of a couple of decades and –since its introduction in clinical practice at the dawn of the millennium– has become an essential tool in the diagnosis and management of small bowel (SB) diseases. At the same time, other fields such as minimally invasive diagnosis of other parts of the gastointestinal (GI) tract has been or is currently explored as future applications of this technology. As a pure imaging modality, VCE ‘suffers’ from lack of additional on-board data that could allow higher diagnostic accurary. This could be either advanced image enhancement or biochemical sensors that could provide relevant info. Furthermore, as VCE clips reading remains manual, it is heavily dependent on the reviewer’s experience. Historically, VCE lesion miss rates have been reported at levels between 6% and 18%. There is also poor agreement on interobserver agreement and subsequent management decisionmaking. The aim of this thesis was to increase the knowledge and to critically evaluate the importance of existing applications as well as exploring and developing new applications to optimize use and diagnostic outcomes of VCE in clinical practice. More specifically, to investigate the correlation between VCE imaging and faecal calprotectin (FC); to develop a model for prediction of VCE results based on FC levels; to investigate and consolidate existing clinical data on the utility of Fujinon Intelligent Chromoendoscopy (FICE) in improving delineation and detection rate for pathological findings in VCE compared to conventional reading; to develop and validate a novel database aiming to provide a reference for research on the development of medical decision support system (MDSS) for VCE; and to develop an approach to capsule localisation and to provide estimations of relative movement of the VCE during its passage through the GI tract. Results of the studies showed that in patients with strong clinical suspicion of SB inflammation and negative (conventional) bidirectional endoscopy, VCE should not be limited to patients with elevated biomarkers only. Moreover, the correlation between FC levels and GI inflammation –as detected by VCE– was moderate and FC=>76 mcg/g may be associated with appreciable SB inflammation on VCE in patients with negative prior diagnostic workup. Furthemore, FICE seems to perform better for pigmented lesions such as angiectasias, both in lesion delineation and detection. However, overall using the three FICE modes did not significantly improve detection rate or the quality of visualization of the most common pathological findings seen on SB VCE. We developed KID, the first database of VCE images and videos with both graphic and semantic annotations, developed specifically for MDSS research. KID provides a platform for data sharing and software development. The experiments detailed are proof-of-principle studies demonstrating the potential for KID to fulfil this role. Moreover, we presented methods for both 2D and 3D localisation of capsules using visual information alone. Such methods are feasible and have potential to be of clinical use.