Deep GI AI: Advancement in Gastrointestinal Cancers Detection Poised to Grow into Startup

Bangkok – The Faculty of Medicine and the Faculty of Engineering, Chulalongkorn University, have developed Deep GI, an artificial intelligence model trained on hundreds of thousands of endoscopic images, to help doctors accurately locate colorectal cancer — matching the expertise of senior GI specialists. Now entering Phase 2, the technology is being expanded to detect and diagnose gastric and bile duct cancers, with plans to spin off into a startup following the Thai FDA approval.

The National Cancer Institute reports that around 140,000 new cancer cases are diagnosed in Thailand each year, with approximately 83,000 deaths — while an average of 227 Thais are losing their lives to cancer every day. In addition, colorectal cancer is reported as the top three cancers over the last 10 years, despite the advancement in diagnostic screening colonoscopy and treatments. One of the main reasons is the disproportion of targeted screening population at 15 million aged above 50 years compared with only 1000 performing endoscopists.

These alarming numbers could decline significantly through early screening and timely treatment. One emerging technology helping doctors detect cancer earlier is artificial intelligence (AI). Among the notable innovations is “Deep GI Innovation” developed by the Center of Excellence for Gastrointestinal Endoscopy (GI), King Chulalongkorn Memorial Hospital, Thai Red Cross Society, and the Faculty of Medicine, Chulalongkorn University.

The Center has successfully used AI to detect colorectal polyp and cancer through endoscopic imaging (Deep GI Phase 1) since 2022 and has now advanced to “Deep GI Phase 2” as the world’s first AI capable of identifying bile duct and gastric cancers in June 2025.

Prof. Dr. Rungsun Rerknimitr, M.D., Assistant to the President for Innovation at Chulalongkorn University, and GI Specialist at the Excellence Center for GI Endoscopy said, “The development of AI for gastrointestinal endoscopy aims to provide general practice doctors with a tool that can accurately detect polyps.”

“Some polyps are difficult to detect because they don’t appear as distinct lumps — some are flat or patch-like, making them easy to miss during endoscopy.  The AI innovation helps highlight abnormal lesions, allowing doctors to target these subtle polyps faster and with greater accuracy,” explained Prof. Dr. Rungsun.

Deep GI Innovation Helps Doctors Pinpoint Cancer Accurately “Deep GI” is a breakthrough innovation jointly developed by the Faculty of Medicine and the Faculty of Engineering, Chulalongkorn University. The name “GI” stands for Gastrointestinal—the digestive system—while “Deep” refers to Deep Learning, a method that allows AI to learn from cancerous and non-cancerous images so it can accurately outline and identify malignant lesions.

“Our team has worked with great dedication to this project.  We used a supervised learning approach to train the AI with hundreds of thousands of images—both cancerous and non-cancerous—from about 500 to 1,000 patients. It took us years before the AI could fully be trained about all the images,” said Prof. Prof. Dr. Rungsun.

According to Prof. Dr. Rungsun, “AI functions like a co-pilot—assisting doctors in pinpointing potential lesions, but AI is not replacing doctors.  Doctors still perform the endoscopy themselves and make the final diagnosis.  In the past, doctors interpreted the images manually, but now, with AI highlighting suspicious areas, physicians can review and confirm the abnormal findings more quickly and accurately.”

Prof. Dr. Rungsun further explained that the Deep GI is installed as an external hardware unit connected to endoscopic equipment, and a computer on the side. The system can retrieve and analyze images directly from any endoscope regardless of the brand or model—and immediately send back annotated visuals to the monitor, annotating suspicious lesions to the performing endoscopist in real time,” Prof. Dr. Rungsun said.

Importantly, Deep GI does not alter the standard endoscopic procedure or add time for patients.

“AI doesn’t lengthen the process, nor does it require special preparation,” Prof. Dr. Rungsun added. “Patients undergoing endoscopy won’t even notice that AI is being used. The procedure remains the same—bowel preparation; sedation, and the endoscopic examination takes about 30 minutes. If a polyp is found, it is removed within three to five minutes, like the standard practice without AI assistant.”

Deep GI Phase 1: Detecting Colorectal Cancer

Colorectal cancer ranks as the third most common cancer among Thais, prompting doctors to focus on screening individuals aged 50 and older. However, the limited number of gastroenterology specialists compared to the country’s population has posed a major challenge. This shortage led to the development of the Deep GI innovation in 2018.

“There are currently about 15 million Thais aged 50 and above nationwide, but only around 1,000 doctors specialize in endoscopic screening — an insufficient ratio.  Therefore we have to use a two steps screening method of colorectal cancer screening by firstly starting with fecal occult blood tests, which help narrow down the target group to about 10 percent — or down to 1.5 million for the next step screening which is colonoscopy. Despite the significant reduction in workload for 1000 performing endoscopists, the proportion of this service remains insufficient to meet Thailand annual screening demands.”

“What we can do is train doctors who are new to endoscopy or just starting their careers to perform procedures as effectively as specialists,” explained Prof. Dr. Rungsun. “AI supports physicians who are still in training or only have a few years of experience, helping them detect colorectal cancer and polyps with greater accuracy and confidence. Our research shows that AI enables these doctors to identify polyps almost as accurately as experts, giving patients confidence that they are receiving the same high-quality care.”

Prof. Dr. Rungsun added that Deep GI will make colorectal cancer screening easier, more accessible, and more comprehensive for at-risk populations.

“It takes about 12 years to train one doctor and become a gastroenterologist, but this skill can be assisted by AI—and its scalability to other doctors has no limits,” he said. “With the necessary hardware or software in place, we can quickly expand screening capacity. The more units of Deep GI systems we deploy, the closer we can get to our national goal of screening 1.5 million Thais aged 50 and above for colorectal cancer.”

Deep GI Expands to Detect Gastric and Bile Duct Cancers

Among 100 patients diagnosed with colorectal cancer, more than 50 percent are detected at a stage that still allows for surgical treatment. However, for gastric and bile duct cancers, fewer than 20 percent of patients are eligible for surgery. This is because these types of cancer are more difficult to diagnose in their early stages and require a higher level of endosocpic expertise during examination.

“Deep GI Phase 2 focuses on detecting cancers of the stomach and bile duct, which are more complex than colorectal cancer. We trained the AI with new datasets, making our team the first in the world to use artificial intelligence for diagnosing gastric and bile duct cancers. Eligible patients include those who already show symptoms or are at risk, and whose physicians recommend endoscopic examination — where the Deep GI system can assist during the procedure.”

“Using AI to detect raised polyps in the colon during endoscopy is relatively straightforward,” explained Prof. Dr. Rungsun. “However, in the stomach, lesions often appear as pale red flat patches that can look normal at first glance. For doctors who are less familiar with this, detection becomes even more difficult. Because these lesions are flat with irregular border, defining their boundaries is challenging — which means the AI used must be far more precise and sophisticated than that for colonic polyp detection.”

“Bile duct cancer is even more difficult to detect. Unlike gastric lesions, the cancerous tissue here doesn’t appear as a patch but rather as small cracks or irregular grooves. Sometimes, even when a biopsy is taken from a suspicious area, it turns out not to be cancerous. The AI assists doctors by targeting the potential lesions as a heatmap—much as how a fighter-jet pilot uses radar to lock onto a precise location. Once the AI highlights an abnormal area in the bile duct or stomach, doctors can accurately examine and precisely biopsy the correct site.”

New Features Added to Deep GI

The advanced AI technology behind Deep GI enables general physicians to detect colorectal cancer with up to 97% accuracy, comparable to that of senior endoscopists. As for the detection of gastric and bile duct cancers, the system is currently undergoing further trials to validate its performance. 

Prof. Dr. Rungsun explained the accuracy of Deep GI, saying, “Generally polyps can be classified into two types — those that are precancerous and those that are not. Currently, determining whether a polyp is likely to become cancerous relies on the physician’s interpretation, which is about 70 percent accurate. In the end, the polyp still needs to be removed and examined under a microscope. However, in the future, the next generation of AI — known as CADx — will be able to analyze and classify polyps directly.”

“In addition to detecting colorectal cancer, Deep GI Phase 1 has been upgraded with diagnostic capabilities, that allow the system to interpret images at a level comparable to specialists — and with greater accuracy than less-experienced endoscopists. While some commercial AI systems offer diagnostic features as optional add-ons, our Deep GI model integrates CADx directly into the package as the standard tool.”

Deep GI Poised for Startup Business

“With artificial intelligence, accuracy depends on the dataset it is trained on. Imported AI systems often lose precision and sensitivity when applied in different populations. When the data comes from Thai patients and is developed by Thai researchers, the results are naturally more accurate for Thais,” explained Prof. Dr. Rungsun.

The professor also mentioned that Deep GI has undergone extensive testing, trials, and research, and has now received approval from the Food and Drug Administration (FDA) for use in general patient screening. The next step, supported by Chulalongkorn University, is to expand its implementation nationwide in public hospitals through the Ministry of Public Health, with funding support from the Board of Investment (BOI). Private hospitals interested in adopting the technology will be able to do so through a Chula Startup company.

“We recently received support from the BOI to launch a pilot project with at least eight hospitals, including King Mongkut Hospital in Phetchaburi, Surin Hospital, Roi Et Hospital, and other hospitals in other provinces,” he added.

Prof. Dr. Rungsun expressed hope that following FDA approval, Chulalongkorn University could engage in B2G partnerships with government agencies such as the Ministry of Public Health, and B2B collaborations with private hospitals interested in adopting the innovation.

“Beyond Deep GI’s current ability to detect cancer quickly and accurately, our next goal is to develop AI that can identify all mature precancerous lesions and determine whether a lesion needs to be removed,” he said.

In closing, Prof.  Dr. Rungsun expressed his vision for the innovation, “We hope Deep GI will serve as a powerful tool for early GI cancer screening—helping prevent disease before it becomes severe, reduce treatment costs, improve patients’ quality of life, and lower Thailand’s overall incidence and mortality rates from gastrointestinal cancers.”

• Facebook
• Twitter
• Line