What is DAMO GRAPE
DAMO GRAPE is the world's first AI model for detecting early gastric cancer based on non-contrast CT, jointly launched by Zhejiang Cancer Hospital and Alibaba DAMO Academy. DAMO GRAPE breaks through the limitations of traditional imaging and, based on deep learning, analyzes non-enhanced CT images to achieve efficient gastric cancer screening. In large-scale clinical research, DAMO GRAPE demonstrated a sensitivity of 85.1% and a specificity of 96.8%, significantly outperforming human radiologists. The model can detect early gastric cancer lesions 6 months in advance, providing a new and efficient means for the early diagnosis and treatment of gastric cancer and expected to greatly improve the survival rate of gastric cancer patients.
Main functions of DAMO GRAPE
- Early gastric cancer screening: DAMO GRAPE analyzes non-enhanced CT images to identify early gastric cancer lesions, significantly improving the early detection rate of gastric cancer.
- Assisted diagnosis: Provides assisted diagnostic support for imaging physicians, helping improve diagnostic accuracy and efficiency and reducing the possibility of missed diagnoses and misdiagnoses.
- Risk assessment: Conducts gastric cancer risk assessment for patients, identifies high-risk groups, and facilitates further confirmatory methods such as gastroscopy.
- Early warning: Detects potential gastric cancer lesions in advance when patients have not yet shown obvious symptoms, buying valuable time for early treatment.
Technical principles of DAMO GRAPE
- Deep learning algorithms: DAMO GRAPE is based on deep learning technology and uses large amounts of CT image data from gastric cancer and non-gastric cancer cases for training to learn the features and patterns of gastric cancer lesions.
- Multi-center dataset: Based on the world's largest multi-center dataset of non-contrast CT images for gastric cancer (6,720 cases), covering data from different regions and different devices, improving the model's generalization ability.
- Image segmentation and classification: The model combines segmentation and classification networks, first segmenting the gastric region in CT images, then performing tumor detection and classification on the segmented region, and outputting a gastric cancer risk score and segmentation mask.
- Feature extraction and recognition: Analyzes subtle changes and patterns in CT images, such as gastric wall thickness and gastric mucosal heterogeneity, to identify early gastric cancer lesions, breaking through the limitations of traditional imaging.
- Grad-CAM visualization: Visualizes the model's decision-making process based on Grad-CAM technology, helping doctors understand the basis of the model's judgments and enhancing the model's interpretability.
Project address of DAMO GRAPE
- Technical paper：https://www.nature.com/articles/s41591-025-03785-6
Application scenarios of DAMO GRAPE
- Large-scale population screening: Conducts initial gastric cancer screening for large populations in physical examination centers and primary hospitals, detecting potential patients in advance and improving the early detection rate of gastric cancer.
- Assisting doctors in diagnosis: Provides radiologists with an assisted diagnostic tool to help identify gastric cancer lesions more accurately, reduce missed diagnoses and misdiagnoses, and improve diagnostic efficiency.
- Monitoring of high-risk populations: Conducts regular screening for residents in areas with high incidence of gastric cancer and high-risk groups such as those with family history or chronic stomach diseases, detecting lesions in advance.
- Early warning intervention: Detects potential gastric cancer lesions in advance when patients have no obvious symptoms, buying time for early treatment and improving patients' survival rates and quality of life.
- Optimization of medical resources: In the hierarchical diagnosis and treatment system, rationally allocate medical resources, improve medical efficiency, and at the same time provide data and tool support for medical research and teaching.