Shulou(Shulou.com)11/24 Report--

The unified diagnosis of multiple cancers has recently ushered in a breakthrough new method. On August 16, Aridamo Institute released a universal model for multi-cancer image analysis, which can detect, segment and diagnose eight major high-risk and fatal cancers, helping to achieve unified diagnosis of multiple cancers and reduce the probability of missed diagnosis.

At present, the medical AI model is powerful enough to complete the disease identification of single organ and assist doctors in diagnosis, but it encounters great challenges in realizing the accurate identification of multiple organs, one is too high false positive problem, the other is there is a certain probability of missed diagnosis, which is especially important for the diagnosis and treatment of patients with multiple cancers. To avoid misdiagnosis and missed diagnosis, radiologists usually detect and diagnose multiple diseases in multiple organs throughout the body. Therefore, doctors urgently need a more efficient unified diagnosis model for multiple cancers in clinical treatment.

In view of the above pain points, the medical AI team of Dharma Hospital, together with the Cancer Prevention and Treatment Center of Sun Yat-sen University, Sichuan Province Cancer Hospital, the First Affiliated Hospital of Zhejiang University, Shengjing Hospital, Guangdong Province People's Hospital and other units, proposed a unified universal model for multi-cancer image analysis (cancerUniT), based on Mask Transformer semantic segmentation, solves the problem that it is difficult to uniformly detect, segment and diagnose multiple tumor images before, and is applicable to eight mainstream high-risk and high-lethal cancers (lung, colorectal, liver, stomach, breast, esophagus, pancreas, kidney) and tumor subtypes in related organs.

The complexity of the multi-cancer problem is mainly reflected in the existence of many associations between organs, malignant tumors and other tumor types. For example, liver cancer and liver cyst are located in the liver, but there are differences in texture and benign and malignant; liver cancer and pancreatic cancer, although similar in shape, are malignant cancers of different organs.

In order to effectively model the differences and similarities between multiple cancers, Dharma Hospital's medical AI team proposed a novel tumor representation learning method with the help of Transformer, representing tumors as semantic Query in Transformer, and establishing semantic hierarchies for tumors and their subtypes in different organs, making the model learning process more effective, improving the consistency of tumor and its subtype predictions, and realizing simultaneous output of segmentation, detection and diagnosis predictions. Therefore, the identification task of clinically complex multi-cancer and multi-tumor can be solved.

In a comparative test of 631 patients, the tumor detection, segmentation, and diagnostic tasks performed better than a single model combination of eight specific organs, with an average sensitivity of 93% and an average specificity of 82% for the detection task.

IEEE Fellow Lu Le, head of the medical AI team of Aridamo Institute, believes that this work realizes "diagnosis of eight most deadly cancers in one call" for the first time with a unified model, simplifying the complexity of AI model while maintaining high sensitivity. This will provide radiologists with comprehensive AI-assisted diagnostic support, especially in clinical scenarios such as cancer recurrence and distant metastasis.

It is understood that the paper achievements of this model have been collected by ICCV2023, and have been applied and tested in many cooperative hospitals such as Shanghai City First People's Hospital.

The medical AI team of Dharma Hospital has been committed to medical imaging and other research for a long time, and is developing cancer diagnosis and treatment technology of the whole process including scale screening, accurate diagnosis, prognostic treatment and response evaluation, covering many important diseases. The team once developed CT image AI auxiliary diagnosis system for new crown pneumonia at the early stage of Xinguan epidemic situation, and was rated as the national advanced group of science and technology anti-epidemic disease by Ministry of Science and Technology.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.