January 15, 2011 - A team of researchers from Stanford Medical School has developed an AI model called MUSK.Combining medical images and text data allows accurate prediction of prognosis and treatment response in cancer patients.
Note: Prognosis (English: prognosis) is a medical term that refers to the estimation of the probable outcome after treatment based on the patient's current condition, combined with what is known about the disease, such as clinical manifestations, laboratory results, imaging tests, etiology, pathology, and patterns of disease, as well as the timing of, and methods of, treatment, and new developments during the course of treatment.
The highlight of the MUSK model is the groundbreaking integration of visual data (e.g., pathology images) and textual data (e.g., medical and clinical records) for a more comprehensive understanding of the patient's condition.
MUSK model in the huge unpairedMultimodalityPre-training on datasets greatly extends its learning range, making it more adaptable and customizable than traditional AI models.
The modeling is done by 50 million pathology images and more than 1 billion medical texts trainingIt can accurately predict patient survival and treatment response for 16 cancer types.
The MUSK model analyzes thousands of data points, including patient demographics and medical history, to more accurately determine which therapies (e.g., immunotherapies) are most effective for an individual patient.
The team said that compared to traditional methods, its accuracy in predicting survival improved by 11 percentage points to 751 TP3T; its accuracy in predicting the suitability of immunotherapy improved from 611 TP3T to 771 TP3T; and its accuracy in predicting the risk of melanoma recurrence within five years improved by 12 percentage points to 831 TP3T.
