When it comes to detecting colorectal cancer in patients, one procedure that has proven helpful is colon capsule endoscopy, during which a pill with a tiny digital camera, a light, a transmitter and a battery in it is swallowed by the patient. These capsules take pictures at up to 30 frames per second, placing a large burden on the medical staff that analyzes the images. With this in mind, Alexander Mamonov and colleagues from the University of Texas at Austin have developed an algorithm that examines each image in the sequence for the typical signs of a polyp and identifies potential candidates for more detailed analysis, according to the MIT Technology Review.
The algorithm puts a focus on protrusions—a key identifier for distinguishing between a polyp and healthy tissues—in an attempt to spot frames that may contain polyps. It works by measuring the curvature of the tissue using a sphere-fitting technique. The radius of the sphere that best fits the tissue fold is then a measure of the curvature, and a threshold curvature above which the frames are flagged for further investigation is set by Mamonov and his colleagues.
When the algorithm was put through its paces on a data set consisting of almost 19,000 images, 230 of which contained polyps, it detected polyps correctly 47% of the time. Since the images are not always straightforward to examine, polyps can be obscured and hard to spot. A better measure may be the ability of the algorithm to spot polyps in the sequence of which they appear, rather than in each frame, says the article. In this measure, the algorithm achieved a recognition rate of 81%.
“While our approach is by no means an ultimate solution to the automated polyp detection problem, the achieved performance makes this work an important step towards a fully automated polyp detection procedure,” Mamonov told the MIT Technology Review.
Also check out:
Ultrasound images from a pill
Share your vision-related news by contacting James Carroll, Senior Web Editor, Vision Systems Design