AI Solution: DEEP:ENDO-ST-01
Core Technology: Upper Endoscopy image processing and uplift and depression pattern analysis technology
Gastric cancer is a malignant tumor of the stomach. It ranked 1st with 12.8% of all cancer incidences in Korea, and the gender ratio is 2:1, ranking 1st among male cancers and 4th among female cancers. (As of 2017, data from the Central Cancer Registry) Early gastric cancer is asymptomatic, so it is important to detect it quickly through examination.
According to Korea's gastric cancer screening recommendations, adults over the age of 40 who have a high incidence of gastric cancer are required to undergo a gastroscopy every two years even if they do not have any symptoms. In gastroscopy, early gastric cancer is detected in the form of elevation or depression, and when advanced, it shows a distinct abnormal appearance such as a mass, an ulcer type, or a diffuse type. We have developed an artificial intelligence product that searches and analyzes the area where the ridge or depression pattern appears in the upper gastrointestinal endoscope. Through this, the convenience of the clinician performing the endoscopy can be improved and errors can be reduced to help diagnose gastric cancer.
AI Core Technology
After homogenizing by applying an image processing technique to each frame of the gastroscope, training was performed through a deep learning algorithm based on Faster-RCNN. Faster-RCNN is a model that shows excellent performance in the field of artificial intelligence object detection, and is used in various fields such as autonomous driving and lesion detection. Faster-RCNN consists of two stages: a Region Proposal Network that proposes a candidate region suspected of being an object of interest in an image, and a Refinement Network that adjusts the size and position of the candidate region and classifies it. In order to deal with various object sizes and shapes, learn the process of converting the size and shape by reflecting the characteristics of the object from the nine types of initial candidate areas (anchors). Since the areas finally detected as objects may overlap each other through this process, the area with the highest probability of being the object is calculated using the Non-Maximum Suppression method by calculating the similarity between the detected areas based on Intersection Over Union (IOU). The other areas except for are removed from the candidate. The trained model automatically extracts features such as the size and location of patterns such as ridges and depressions from the input gastroscopy image and finally displays them in the form of a red box.
AI Solution: DEEP:ENDO-CL-01
Core Technology: Colon endoscopy image processing and uplift and depression pattern analysis technology
Colon cancer is a malignant tumor that occurs in the colon. It ranks second with 12.1% of all cancer incidences in Korea, and the gender ratio is 1.5:1, which is more common among men. (As of 2017, data from the Central Cancer Registry) In the early stages, most of colon cancer has no symptoms, and when symptoms appear, there are many cases that have already progressed considerably. Therefore, if colon cancer is detected and treated at an early stage, the treatment results are very good. Therefore, it is very effective to find the polyp in the adenoma stage through examination and remove it with an endoscope.
The national cancer screening program for colorectal cancer in Korea requires that all men and women 50 years of age or older perform a colonoscopy every year if there is an abnormality in the differential occult blood test, and the National Cancer Center and the Korean Colon and Anus Society report 5~ It is recommended to perform a colonoscopy once every 10 years. Colorectal cancer can be observed not only in the form of a polyp, but also in a variety of appearances such as ridges, depressions, and bleeding in the colon wall, so it is essential to detect and analyze these lesions without missing them.
We have developed an artificial intelligence product that searches and analyzes areas where ridges, depressions, and color changes appear in colonoscopy image data. This increases the convenience of the clinician performing colonoscopy and lowers errors, which can help diagnose colon cancer.
AI Core Technology
This product is a solution that detects the pattern in the image by learning features such as the size and location of ridges and depressions in colonoscopy images using artificial intelligence technology. For each frame of the colonoscopy, an image processing technique was applied and uniformed, and then training was performed through a deep learning algorithm based on Faster-RCNN. The trained model automatically extracts features such as the size and location of patterns such as ridges and depressions from the input colonoscopy image and finally displays them in the form of red boxes.