Chinese scientists achieve breakthrough in early detection of ‘king cancer’ that killed Steve Jobs

• AI scientists and clinical researchers have worked together to develop a new early screening method to detect pancreatic cancer

• It could help save thousands of lives every year, with the difficulty in diagnosing pancreatic cancer making it one of the deadliest cancers

An artificial intelligence tool developed by Chinese scientists has led to a breakthrough in early-stage screening of one of the most fatal cancers.

Pancreatic cancer, often called the “king of cancers”, has an average five-year survival rate of less than 10 per cent. It killed Apple co-founder Steve Jobs in 2011, and more recently, it was the cause of death last month of Wu Zunyou, chief scientist at the Chinese Centre for Disease Control and Prevention.

One of the main reasons pancreatic cancer has such a high death rate is the difficulty in early detection. It is rarely found in its early stages, when the chance of curing it is at its greatest. That is because it often does not cause symptoms until it has spread to other organs, according to the Mayo Clinic.

But the early screening model, developed jointly by AI scientists from tech firm Alibaba Group’s DAMO Academy and clinical researchers from hospitals including the Shanghai Institution of Pancreatic Diseases, has shown promising results. Alibaba is the owner of the South China Morning Post.

The model combines a non-contrast computed tomography (CAT) scan with an AI algorithm. In a paper published by the peer-reviewed journal Nature Medicine on Monday, the team said the specificity of the early screening model reached 99.9 per cent, implying there is only one false-positive case in every 1,000 tests.

Meanwhile, its sensitivity, or ability to detect pancreatic tumours, could reach 92.9 per cent, beating mean radiologist performance by 34.1 per cent.

One of the paper’s reviewers, Li Ruijiang, an associate professor of radiation oncology at the Stanford School of Medicine, said the work represented “an important step in the right direction for pancreatic cancer screening”.

Yet AI-based imaging applications have not been granted approval by Chinese authorities, according to a doctor from the Cancer Hospital at the Chinese Academy of Medical Sciences who declined to be named. So despite its impressive initial results, there is still a long way to go before the technology could be used in clinical practice.

The early screening model developed by the team is tailored for pancreatic ductal adenocarcinoma (PDAC), the most common subtype of pancreatic cancer, which accounts for over 95 per cent of all cases. PDAC causes around 466,000 deaths per year worldwide.

In the United States, pancreatic cancer is now the fourth leading cause of cancer-related deaths for both men and women, and trends indicate that it will be the second leading cause of cancer death in the country by 2030.

Research published in April by the US National Institutes of Health (NIH), which looked at trends in age-standardised cancer incidence, survival and mortality rates between 2000 and 2019, found that death rates from pancreatic cancer have been steadily increasing – 0.2 per cent every year from 2006 to 2019.

Early or incidental detection can significantly improve a patient’s chance of survival. Studies have shown that high-risk patients who had PDAC detected in early screening have a median overall survival of 9.8 years, while those with late diagnoses have a median survival of 1.5 years.

However, there is a lack of effective and easily accessible screening technology for the general population.

With the prevalence of pancreatic cancer relatively low – there are fewer than 13 cases per 100,000 – the use of expensive contrast-enhanced CAT scans across the general population is uneconomic.

“On the other hand, existing early screening tools for pancreatic cancer are generally poor in accuracy, leading to many cases of misdiagnosis and unnecessary panic,” said lead author Cao Kai from the Shanghai Institution of Pancreatic Diseases in an interview with mainland media website Zhishifenzi.

It was during a conversation last year between Cao and Lu Le, the leader of DAMO Academy’s medical team, that they came up with the idea of using AI to assist in early cancer screening.

The pair quickly took action and, together with more than 10 prestigious medical institutions, they initiated a research project aimed at developing a technology that would combine non-contrast CAT scans, which are widely used in medical facilities and hospitals, with AI to create a model suitable for large-scale pancreatic cancer screening.

Apple co-founder Steve Jobs died at the age of 56 from pancreatic cancer, which is notoriously difficult to detect. Photo: AP

Their brainchild was an algorithm for “pancreatic cancer detection with artificial intelligence” – known as PANDA. It was trained based on more than 3,200 image sets from a high-volume pancreatic cancer institution in China, about 70 per cent of which stemmed from patients with a pancreatic lesion.

Thanks to the large data set, meticulous data processing and innovative training strategy design, PANDA was trained as a highly perceptive AI imaging expert.

Researchers at DAMO Academy discovered that subtle density differences in non-contrast CAT scans, which may be difficult to detect with the naked eye, can be picked up by AI.

When applying PANDA to real-world clinical scenarios involving 20,530 patients to validate its accuracy, the researchers found that the AI tool could achieve impressive sensitivity of up to 92.9 per cent and a specificity of 99.9 per cent.

According to information provided by Alibaba Cloud, the PANDA model has been used more than 500,000 times in settings including hospitals and medical examinations, and has detected multiple cases of early-stage pancreatic cancer that were previously missed.

“The accuracy metrics of the PANDA algorithm are superior to those of several acknowledged screening methods,” German clinical expert Joerg Kleeff and his colleague wrote in a comment piece published in the same Nature issue.

However, they also warned that further assessment was needed before the AI-based screening could become widespread practice.

They pointed out that any potential screening method for pancreatic cancer should detect early stages such as “T1 lesions”, which are smaller than 2cm (0.79 inches) in diameter, but the AI model from China did not report specificity and predictive values for this subgroup.

“The value of any screening method for cancer lies in reducing all-cause mortality. The study was of retrospective design and so could not assess the effect of screening on the mortality of included patients,” they said, adding that AI-based screening should be evaluated with the same rigour as conventional screening.

“This AI model is still at the early stage and warrants more validation efforts,” the doctor at the Chinese Academy of Medical Sciences said. He added that given the low prevalence of pancreatic cancer, the demand for this AI tool would be limited.

Source