Cure｜AI-Powered Early Screening Brings New Progress in Fighting King of Cancers—Pancreatic Cancer (AI Translation)

解药｜AI早筛“癌中之王”胰腺癌有新进展

Pancreatic cancer, often dubbed the "king of cancers" due to its aggressive nature and poor prognosis, remains difficult to diagnose at an early stage. Its deep anatomical location and lack of early symptoms contribute to a less than 5% early detection rate, with about 60% of patients already experiencing metastasis by the time of their first diagnosis. The five-year survival rate stands at only 7.2%-9%. Most cases present late, with symptoms like abdominal pain or jaundice prompting imaging studies that lead to a diagnosis. [para. 1][para. 2][para. 3]

In response to these challenges, artificial intelligence (AI) has been introduced as a promising screening tool. In November 2024, Ningbo University Affiliated People's Hospital launched a project using the "DAMO PANDA" AI system, developed by Alibaba DAMO Academy, to assist in early pancreatic cancer screening via non-contrast CT scans. This AI system analyzes routine abdominal and chest CT images, detecting subtle lesions with a sensitivity of 92.9% and specificity of 99.9%, according to a study published in Nature Medicine. By March 2025, over 30,000 people had been screened at the hospital, identifying seven cases of early pancreatic cancer, none of whom had typical pancreatic cancer symptoms. Early surgical intervention can significantly improve five-year survival rates from below 10% to 30-40%. [para. 4][para. 5][para. 6]

China faces a high burden of pancreatic cancer. In 2022, there were approximately 118,700 new cases annually, ranking it tenth among malignancies. The disease is characterized by high incidence, mortality, relapse, and metastasis rates, but low early diagnosis, surgical resection, and survival rates. Surgery is the preferred treatment; however, only about 20% of patients at this hospital are surgical candidates, with most presenting at advanced stages. For those who cannot undergo surgery, options like chemotherapy may offer limited benefit, and the disease often progresses rapidly—late-stage patients not responding to therapy may have a survival time of only 3-6 months. Consequently, early detection is critical for improving outcomes. [para. 7][para. 8][para. 9][para. 10]

Traditional diagnostic tools—contrast CT, MRI, and endoscopic ultrasound—are often invasive, expensive, or technologically demanding, and are not widely available at lower-level hospitals. In contrast, non-contrast CT is more affordable (about 200-300 yuan per scan versus 600-900 yuan for enhanced CT), widely accessible, and suitable for repeated screening, especially for high-risk individuals like those with family history. AI now enables such scans to effectively identify potential pancreatic cancers, reducing barriers to early detection. [para. 11][para. 12][para. 13][para. 14][para. 15]

Ideal screening tools must be both low-cost and highly accurate, with minimal false positives and negatives. In a study of 20,530 patients, the DAMO PANDA system demonstrated substantial improvements over radiologists, increasing sensitivity by 34.1% and specificity by 6.3%. Despite these advances, the system is still under clinical evaluation and has not yet received regulatory approval. It could benefit from incorporating multimodal data (such as blood tests and family history) to further improve accuracy. [para. 16][para. 17][para. 18][para. 19]

DAMO’s broader vision is for "one scan, multiple screenings"—using a single CT scan to screen for various cancers and chronic diseases, potentially making widespread, opportunistic screening possible for entire populations. This approach contrasts global molecular diagnostics, such as GRAIL's Galleri test—which screens for 50 cancers from a blood sample but costs around $1,000 per test and has not yet received regulatory approval. Both strategies face challenges in demonstrating sufficient sensitivity, specificity, and population benefit for multi-disease screening. [para. 20][para. 21][para. 22][para. 23][para. 24][para. 25]