- Approach skips bowel prep, targets early detection
- Study shows higher sensitivity than clinicians
Alibaba DAMO Academy has developed an AI model for colorectal cancer screening that uses routine CT scans, offering a method that does not require bowel preparation and can be conducted without additional procedures for patients.
The model, known as DAMO Coca, was created in collaboration with Guangdong Provincial People’s Hospital and other institutions.
It marks the third cancer screening model released by the research unit, following earlier systems targeting pancreatic and gastric cancers, and signals progress in a broader “CT scan plus AI” approach to multi-cancer detection.
Colorectal cancer is the second leading cause of cancer-related deaths globally, with rising incidence among people under 30.
Early detection can lift five-year survival rates above 90%, compared with about 14% for late-stage cases.
Yet conventional screening methods, including stool tests and colonoscopy, often require invasive preparation, contributing to low uptake rates.
DAMO Coca applies AI to standard, non-contrast CT scans that are already widely used in health checks and clinical assessments.
While such scans generate large volumes of imaging data annually, the presence of intestinal contents has historically made interpretation difficult.
The system uses a two-stage deep learning architecture combined with hybrid supervised learning to identify early tumors smaller than three centimeters.
The AI model is trained to segment intestinal structures, filter out interference and detect suspicious lesions.
In a study involving 27,000 CT scans, the model identified five previously missed colorectal cancer cases. It achieved a sensitivity of 86.6% and specificity of 99.8%, according to findings published in Annals of Oncology.
Compared with 10 radiologists of varying experience, the DAMO Coca’s sensitivity was 20.4 percentage points higher, with notable gains in areas such as the sigmoid colon and rectum where lesions are often overlooked.
When used as an assistive tool, it improved doctors’ sensitivity and specificity by 14.5% and 3.1%, respectively.
“AI tools can effectively help doctors address the challenge of missed intestinal lesions,” said Liu Zaiyi, head of radiology at Guangdong Provincial People’s Hospital. “Future deployment across more institutions and large-scale prospective studies will be needed to generate high-quality evidence.”
Alibaba DAMO Academy has been working on medical AI since 2017, focusing on detecting subtle abnormalities in CT scans. Its previous models include DAMO Panda for pancreatic cancer and DAMO Grape for gastric cancer.
Zhang Ling, a senior algorithm expert at the academy, said the team has made progress across five major digestive system cancers—pancreatic, gastric, colorectal, liver and esophageal—and is continuing research into screening for breast and kidney cancers.
