Chinese scientists pave the way for early pancreatic cancer detection via AI revolution

126

27th November 2023 – (Beijing) Chinese scientists have devised an artificial intelligence (AI) tool that promises a breakthrough in early-stage screening of deadly pancreatic cancer. The disease possesses a grim five-year survival rate of less than 10%, mainly due to the challenges in early detection.

The formidable death toll from pancreatic cancer stems primarily from its elusive nature in the early stages. Often asymptomatic until it metastasises to other organs, the disease is seldom detected when treatment could be most effective. However, a recent development by AI experts from Alibaba Group’s DAMO Academy and clinical researchers from institutions like the Shanghai Institution of Pancreatic Diseases could be a gamechanger.

The team’s early screening model, which blends non-contrast computed tomography (CAT) scans with an AI algorithm, has demonstrated impressive results. According to the peer-reviewed journal Nature Medicine, the model’s specificity achieved a staggering 99.9%, suggesting only one false-positive case for every 1,000 tests. Furthermore, its sensitivity, or its capacity to identify pancreatic tumours, was 92.9%, outperforming the average radiologist performance by 34.1%.

These findings were hailed as “an important step in the right direction for pancreatic cancer screening” by Li Ruijiang, an associate professor of radiation oncology at Stanford School of Medicine. However, it’s worth noting that AI-based imaging applications are yet to receive approval from Chinese authorities. Therefore, despite the encouraging initial outcomes, there is considerable groundwork to be covered before this technology can be implemented in clinical practice.

The model is specifically designed for early detection of pancreatic ductal adenocarcinoma (PDAC), the most prevalent subtype of pancreatic cancer, accounting for over 95% of all cases. PDAC is responsible for approximately 466,000 deaths globally each year. The research reveals that early detection can significantly prolong a patient’s life. Patients with early-detected PDAC have a median survival of 9.8 years, compared to only 1.5 years for those diagnosed late. However, effective and readily available screening technology for the general public is currently lacking.

The low prevalence of pancreatic cancer, fewer than 13 cases per 100,000, makes the use of costly contrast-enhanced CAT scans for mass screening economically unviable. Existing early detection tools for pancreatic cancer often lack accuracy, leading to misdiagnoses and unnecessary distress, as stated by lead author Cao Kai from the Shanghai Institution of Pancreatic Diseases.

The idea of integrating AI into early cancer screening emerged during a discussion between Cao and Lu Le, the head of DAMO Academy’s medical team. They swiftly initiated a research project with over ten prestigious medical institutions to develop a technology combining non-contrast CAT scans with AI, aiming to create a model suitable for large-scale pancreatic cancer screening.

Their creation, an algorithm named PANDA (Pancreatic Cancer Detection with Artificial Intelligence), was trained using more than 3,200 image sets from a high-volume pancreatic cancer institution in China. Approximately 70% of these images were derived from patients with a pancreatic lesion.

The researchers at DAMO Academy found that subtle density differences in non-contrast CAT scans, often overlooked by the human eye, could be recognised by AI. Upon testing PANDA in real-world clinical scenarios involving over 20,000 patients, the AI tool demonstrated promising sensitivity of up to 92.9% and a specificity of 99.9%.

Alibaba Cloud confirms that the PANDA model has been deployed over 500,000 times in various settings, including hospitals and medical examinations. It has successfully detected multiple cases of early-stage pancreatic cancer that had previously gone unnoticed.

Despite these promising results, experts advise caution before widespread implementation of this AI-based screening. They stress the need for further assessment and rigorous evaluation, similar to that applied to conventional screening methods. The AI model is still in its nascent stages and requires more validation efforts, especially considering the low prevalence of pancreatic cancer.