Introduction Upper gastrointestinal (GI) endoscopy is invasive, incurs risks and most findings are not relevant to the symptoms. Yet 1% of the population undergo upper GI endoscopy per annum and the Be Clear on Cancercampaign expects a 52% increase in demand. Several studies suggest that capsule endoscopy (CE) could be used to examine the upper GI tract, but lack of control and difficulties locating landmarks could increase the chance of missed pathology. We aimed to see if it were possible to map the gastric mucosal surface using mosaicking algorithms to stitch together overlapping CE video frames: a technique which could be used to inform clinicians about completeness of examination.
Method An algorithm for image mosaicking from CE video frames was developed based on the feature selection method SURF (speeded up robust features). This ’blob detection’ method recognises local regions of an image which are distinguishable by properties such as brightness and matches them in overlapping images. SURF is fast to compute, adapts to variations in rotation and scale and therefore is potentially suitable for analysing CE images.
In order to establish that there were enough features for image stitching in the CE images we performed a control test: a single CE image from a forward viewing Mirocam Navi (Intromedic, Seoul, South Korea) was divided up into overlapping subsections. Each subsection was separately submitted to a random transformation of scaling, rotation and translation. These images were then given as an input sequence to the algorithm in order to be stitched back together. A further test of the image stitching algorithm was performed using the Capsocam (Capsovision, CA, USA) capsule, which incorporates four 90° side viewing cameras and is therefore advantageous for mapping because of its 360Opanoramic view.
Results Using control images from human stomachs we established that there are sufficient features in gastric CE images to allow feature based image mosaicking. This applied to imaging obtained both from both forward viewing and panoramic view devices. Finally, we found that the success of the image mosaicking depended on the movement of the capsule, and whether there were sufficient overlapping features from frame-to-frame for stitching images together.
Conclusion Our data show that there are sufficient features to stitch together CE video frames confirming the possibility of developing gastric surface mapping using image mosaicking techniques. A degree of control may be necessary in order to achieve a ‘slow sweep’ over the mucosal surface in order to maximise the acquisition of overlapping features.
Disclosure of interest None Declared.
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.