Iris Coleman
February 4, 2025 22:46
Stanford University Researchers developed MUSK, an AI model that improves cancer diagnosis and treatment through multimodal data processing, making the existing model better than accuracy and prediction.
Researchers at Stanford University have unveiled a groundbreaking AI model called MUSK (MUSK), which aims to simplify cancer diagnosis and personalize treatment plans. This innovative model was set to develop precise oncology by adjusting a treatment plan based on its own patient data as reported by NVIDIA.
Multimodal data integration
The MUSK uses a two -stage multimodal transformer model to handle both clinical text data and pathology images. This approach provides improved clinical insights because the model can identify patterns that can not be detected immediately by medical professionals. This model can first improve this understanding by improving this understanding through the image text data that is learned and paired in a large amount of paired data, and can recognize cancer types and biomarkers and propose effective treatment.
Unprecedented data processing
The AI model was pre -treated in advance using a real data set consisting of 50 million pathological images of 11,577 patients and more than 1 billion pathology. This extensive pre -honey was carried out over 10 days using 64 NVIDIA V100 Tenser Core GPUs and emphasized the capacity of a model that can efficiently process large data.
Excellent performance of diagnosis
When evaluated in 23 pathology benchmarks, the MUSK surpassed the existing AI model by effectively matching the pathology image and the medical text. It also showed 73% accuracy in pathology -related questions, such as identifying cancer and predicting biomarker presence.
Improve cancer detection
The MUSK has improved the detection and classification of various cancer subcontractors, including breast cancer, lungs and colon rectal cancer, up to 10%. It also showed 83%accuracy to detect breast cancer biomarkers and predicted cancer survival with a 75%success rate. This model surpasses standard clinical biomarkers, and usually only 60-65%accuracy.
Future prospect
The team plans to verify models in various patient population and clinical environments aimed at approval of regulatory clinical trials. It also explores MUSK’s applications for other data types such as radiation, images and genomic data to further improve the diagnostic function.
Researchers, including installation guidelines and model evaluation codes, are provided by Github and provide resources for further exploration and development in the medical AI field.
Image Source: Shutter Stock