AI Helps ferret out the symptomatic polyps of Colorectal Cancer, A medical inspection project collaborated by Dept. of Computer Science and Information Engineering of Chung Cheng University with Chiayi Christian Hospital

Published on Jan 28 2019

The idea of combining Artificial intelligence (AI) with medicine has never been so popular. Today, AI can also be used to assist physicians to detect possible cancerous colorectal polyps promptly. Dr. Wei-Min Liu, the Associate Professor of Dept. of Computer Science and Information Engineering at the National Chung Cheng University (CCU), lately has worked with Chiayi Christian Hospital (CYCH) to develop a computer-based automatic detection system by using AI, Big Data Analytics and Deep Learning techniques. This system with its 95% accuracy, will not only help doctors to detect and classify the colorectal polyps, it is also expected to make doctors’ jobs more efficient in the diagnosing process. More, when performing the computerized tomography (CT) scans, a similar AI will help doctors to identify and trace the boundaries of organs, from several to hundreds of scan images, much faster.

Whenever walk into the hospital, it seems that there are always many patients waiting in line to do colonoscopy for the examination or removal of potential cancerous polyps. In the past 11 years, colorectal cancer has been consecutively ranked No.1 among the top ten cancers that occur in Taiwan. Its prevalence rate is significantly high and the colonoscopy is a common practice used for colorectal cancer prevention. However, the inspection practice is not perfect as the symptoms have sometimes got misdiagnosed. Owing to the doctors who only rely on the experience to locate and identify the polyps with their naked eyes, and the inevitable disturbance such as water and foam found inside colon and rectum, the colonoscopy inspection does have difficulties to be faultless. On top of that, most of the polyps research today hardly take into account the credit of prompt detection. Therefore, Professor Liu and his graduate student Wei-Ting Xiao have applied AI technology, in cooperation with Dr. Li-Jen Chang from the Division of Gastroenterology of CYCH, to develop a new system in the hope to allow doctors to discover the problematic polyps more efficiently and accurately.

Colorectal polyps can be divided into neoplastic and non-neoplastic, two different types of polyps. "Adenomatous polyps are more commonly seen. It is recommended to remove those polyps as they may turn into colorectal cancer in a few years." Professor Liu said. After the application of human clinical trial in CYCH has been approved by the authority, the colonoscopy video images and other clinical data provided by the hospital were sent to AI-tech team of Professor Liu. With the doctors’ assistance in marking down the locations of polyps on each image, Liu’s team then used all that information to train the AI to identify the targets automatically. Currently, during the colonoscopy exam the system has allowed doctors to have real-time images on the computer screen, and the polyps are marked and classified automatically, with an accuracy of 95%.

In addition to working with CYCH on the project of colorectal polyp detection, Professor Liu has also introduced AI technology in the process of radiation therapy at the Division of Radiation Therapy of Dalin Tzu Chi Hospital (DTCH). Usually, before the proceeding of radiotherapy, each cancer patient needs to perform a CT scan, then the appropriate dosage of radiation for the treatment will be calculated accordingly by the doctors. Professor Liu pointed out that since every CT scan would produce hundreds of images and there are many patients doing CT scans every day; it is very labor-dependent and time-consuming for doctors to identify the cancerous tissues and surrounding organs to work out the right dosage for each patient particularly. It is indeed a big workload.

However, the AI system that has been developing by Professor Liu’s team will help doctors reduce the time spent on manual delineation significantly. The new system can automatically trace the boundaries of some organs on the images, allowing an at-least-one-hour job to be done in only 10 minutes. "Let the system do the incipient work in marking the approximate boundaries of each organ first, then the doctors would only need a small amount of time to refine and adjust the results of AI later." Professor Liu explained. Thus, the doctors will be able to use their time more constructively, and the patients may be admitted into the treatment sooner. Both doctors and patients can be benefited from the system.