On August 16, 2023, DAMO Academy, under the Alibaba umbrella, announced a universal model for multi-cancer image analysis that can detect, segment, and diagnose eight major types of cancer with high morbidity and fatality rates, helping to achieve simultaneous diagnosis of multiple cancers and reduce the probability of missed diagnosis.
AI models in the medical field are capable of identifying diseases in single organs and support physicians in their diagnosis, but it has been difficult to correctly identify them in multiple organs. Therefore, there was a need for a model that could efficiently diagnose multiple cancers together, and DAMO Academia's medical AI team, in collaboration with the Tumor Prevention and Treatment Center of Zhongshan University, Sichuan Tumor Hospital, the First Hospital Affiliated to Zhejiang University, and Guangdong People's Hospital in China, built a model that can diagnose multiple cancers from images ( cancerUniT) to diagnose multiple cancers from images. Previously, it was not possible to detect, segment, and diagnose from images of various tumors at once, but with the model built on the semantic segmentation model Mask Transformer, eight types of cancer with high morbidity and fatality rates (lung, colon, liver, stomach, breast, esophagus, pancreas, kidney) and various types of tumors in related organs.
To model the differences and similarities between the various cancers, the Dharma Institute Medical AI team developed a way to learn to display tumors using the deep learning model Transformer. The idea is to display tumors as Transformer queries and build semantic layers for different organ tumors and their types. In tests on 631 patients, the performance of tumor detection, segmentation, and diagnosis exceeded that of the combination of single models for the eight organs. The average sensitivity and specificity of detection reached 93% and 82%, respectively.
The DAMO Academy AI team states that for the first time, the unified model allows for "a single diagnosis of eight highly lethal cancers," simplifying the complex AI model while maintaining high sensitivity. The paper on this model was accepted for publication at the international computer vision conference ICCV 2023 and is currently being tested at several hospitals in China, including the First People's Hospital of Shanghai.