GRAPE is Alibaba’s AI-driven breakthrough in non-invasive gastric cancer detection. It redefines early diagnosis through precision CT scan analysis, beating human radiologists and providing a scalable, patient-friendly alternative to endoscopy—especially in countries where cancer often goes undetected due to limited screening.
Alibaba’s GRAPE AI Detects Gastric Cancer Early from CT Scans — Outperforms Radiologists
Gastric cancer is one of the deadliest forms of cancer worldwide, primarily due to its silent progression and delayed diagnosis. But a new innovation from Alibaba’s Damo Academy—the GRAPE AI model—is poised to change that. This AI-based system can now detect gastric cancer, including its early stages, using only non-invasive CT scans. And the results are astonishing: it’s more accurate than human radiologists.
This landmark model, named GRAPE (Gastric cancer Risk Assessment Procedure), was co-developed with the prestigious Zhejiang Cancer Hospital and has already passed rigorous clinical studies published in Nature Medicine. It boasts a sensitivity of 85.1% and a specificity of 96.8%, significantly outperforming traditional methods that require invasive endoscopy.
Background: What Is GRAPE AI and Why It Matters
GRAPE AI is not just another algorithm—it represents a profound shift in medical diagnostics. Developed by Alibaba’s Damo Academy in collaboration with the Zhejiang Cancer Hospital, it’s been trained on the largest known dataset of gastric cancer CT images. That’s a major leap, as plain CT scans were previously considered ineffective for stomach imaging due to anatomical variability.
What sets GRAPE apart is its ability to recognize tiny, subtle patterns that even experienced radiologists might miss. With the help of machine learning and deep image segmentation techniques, GRAPE scans three-dimensional CT images and highlights early indicators of tumors—before they grow too advanced.
This innovation addresses a major gap in China’s healthcare system, where fewer than 30% of patients opt for endoscopy—a painful, camera-based diagnostic procedure. GRAPE eliminates this barrier with a completely non-invasive, fast, and scalable method.
How GRAPE Works: AI That Outperforms Experts
Trained on thousands of expertly labeled images, GRAPE’s neural network is tuned to catch the slightest anomalies. In one real-world case cited by Alibaba, GRAPE flagged a tumor six months before it was detected by human radiologists. This could mean the difference between life and death for many patients.
Its performance metrics—85.1% sensitivity (ability to correctly detect cancer) and 96.8% specificity (ability to avoid false positives)—surpass most standard screening tools. In clinical trials across 20 hospitals involving nearly 100,000 participants, the AI demonstrated not just technical excellence but also real-world impact.
Why This Matters Globally—Especially in Asia
In countries like China, Pakistan, and India, stomach cancer remains a leading cause of cancer-related death. Lack of routine screening, cultural resistance to invasive diagnostics, and limited healthcare access have compounded the issue. GRAPE is emerging as a scalable, accessible tool to bridge that gap.
GRAPE’s rollout will begin in Zhejiang and Anhui provinces, targeting large-scale screening campaigns in urban and semi-rural areas. This paves the way for expansion into other regions where stomach cancer is a significant burden.
AI Beyond GRAPE: The Rise of Damo Academy in Healthcare
GRAPE builds on the success of another Alibaba innovation—Damo Panda, an FDA-recognized AI for pancreatic cancer screening. These tools represent a new era of medical AI, one where machine precision augments human expertise, improving outcomes without replacing the physician.
With each model, Alibaba Damo Academy is cementing its position as a global AI leader not just in tech, but in life-saving medical science.
Expert Opinions: “This Could Save Millions”
“This model offers the best of both worlds—accuracy and comfort,” said Dr. Cheng Xiangdong, gastric surgeon at Zhejiang Cancer Hospital. “It overcomes patient resistance, ensures early detection, and integrates seamlessly into clinical workflows.”
Doctors in the field emphasize that early-stage detection is where GRAPE shines. As the model gains FDA-equivalent approvals internationally, it could revolutionize screening strategies in countries like Pakistan, Japan, South Korea, and India—where the prevalence is high and healthcare infrastructure varies.
Explore how AI is transforming other fields of medical diagnostics in our deep-dive:
ArkPlus vs Google Chest AI: The Race to Diagnose From X-Rays
Alibaba’s GRAPE AI is a pioneering deep learning model trained to detect gastric cancer at an early stage using standard CT scans. Traditionally, CT imaging was not considered sufficient for accurate diagnosis of gastric tumors due to organ complexity and variability. However, GRAPE overcomes these challenges by leveraging a massive dataset of annotated cancer images and advanced segmentation algorithms. This allows it to identify early-stage cancers—often invisible to the human eye—with unprecedented accuracy.
GRAPE was trained on tens of thousands of CT scans curated from China’s top oncology centers, including Zhejiang Cancer Hospital. Using 3D convolutional neural networks, the model interprets multiple cross-sectional slices and reconstructs risk zones in high resolution. This approach significantly reduces false negatives, enabling the model to pinpoint lesions before they evolve into aggressive forms of cancer. The training involved rigorous cross-validation and clinician oversight to maintain clinical relevance and ethical integrity.
Peer-reviewed and published in Nature Medicine, GRAPE’s performance metrics—85.1% sensitivity and 96.8% specificity—were validated through multicenter trials involving over 94,000 subjects across 20 hospitals. Experts confirm that this model outperforms many conventional diagnostics and sets a new benchmark in AI-assisted radiology. The model's early detection capability has already saved lives by identifying tumors months before human doctors noticed abnormalities in the same scans.
This technology is especially promising for developing nations where access to specialized cancer screening like endoscopy is limited. GRAPE’s non-invasive approach, requiring only CT infrastructure and cloud-based processing, makes it scalable across rural and urban hospitals alike. Experts predict that models like GRAPE will form the backbone of AI-first diagnostic networks in Asia, Africa, and the Middle East in the next five years.