Machine Learning - GastroView project
We find and tag anomalies in capsule endoscopy images
play video
Due to Deep Learning algorithms implemented in GastroView, it is possible to shorten time of examination and analysis of image received from capsule endoscopy by maximum of 70%.
01
How does it works?
Step 1
During capsule endoscopy a capsule size of a large pill containing two wide-angle cameras, is sliding down the digestive tract and automatically recording insides of internal organs. Images recorded by cameras after being magnified eight times allows to notice changes approximate to 0,1 cm.
Step 2
During 8 hours cameras take about 100 000 pictures. They are encoded and wirelessly transmitted to data recorder that the patient wears during the examination. Received video is later analyzed by a doctor.
Step 3
Usual time of analysis and evaluation of the results of capsule endoscopy depends on diagnosed disease as well as on experience of the doctor and takes many hours. GastroView software is going to reduce that time by maximum of 70% and due to that costs reduction by approximately 50%. Simultaneously, the accuracy of performed data analysis increases as well as the number of diagnosed anomalies.
02
Technology
GastroView software uses algorithms based on Deep Learning methods performed with nVidia graphic cards processors. Thanks to them we can speed up the calculations on Neuron Networks - the are up to ten times faster comparing to methods based only on classical CPUs. Applying advanced methods allows to automatically localize polyps, bleeding areas as well as other anomalies located anywhere in digestive track.
Automatic classification of anomalies in video data of digestive track indicates a need for creating computer models that will make an interpretation of image in the form of matrix of numbers in search for presence of complex visual elements of significant diversity such as polyps. In order to fulfill such a task we use Convolutional Neural Networks, that for now are known as the greatest methods of analysis and interpretation of complex images and visual objects.
object recognition variant
increasingly complex features
simple inputs
03
Achievements
CTA have been invited to the conference "Fast track to the innovativeness of Polish companies" organized by the National Centre for Research and Development. The event promoted Polish companies that can receive substantial grants to conduct research and development and as a result of the implementation of your product or service to the mass market. From the right Polish Minister of Science and Higher Education - Jaroslaw Gowin and Member of the Management Board CTA Sp. Zoo. - David Jereczek.
partners