Artificial intelligence combined with DNA studies reveals precancerous lesions that are invisible to traditional techniques. However, many do not actually lead to cancer. There are still multiple complex challenges ahead for the early detection of pancreatic cancer.
Up until now, PanINs (pancreatic intraepithelial neoplasias) have been the trickiest precancerous lesions to find, because they are embedded deep within the pancreas, which in itself is very awkward to access. This explains why cancers of this vital gland — the pancreas produces insulin and other hormones — are still, all too often, diagnosed late, when there are few therapies available to fight them. But a technique developed by researchers at The Sol Goldman Pancreatic Cancer Research Center, at Johns Hopkins University in Baltimore (USA), might change this, as reported by the scientific journal Nature.
Early diagnosis of pancreatic cancer using artificial intelligence
This method brings together the best technologies of our time, from artificial intelligence to the study of the genetic code (genomics), producing a 3D reconstruction of the internal structure of the pancreas, with ultra-high, single-cell resolution. In this way, the American researchers were able to create the most detailed map of precancerous lesions in the pancreas to date, laying the foundation for future early diagnosis of pancreatic ductal adenocarcinoma — the most common, and most fearsome, form of pancreatic cancer — and other types of cancer.
To achieve this, the scientists sliced 46 pancreatic tissue specimens into thousands of sections, stained them using traditional techniques and mounted them onto the same number of sequential slides. Finally, they trained an artificial intelligence system called CODA to analyse these slides and provide a 3D reconstruction of the average internal structure of a pancreas.
Unexpected findings
The researchers were surprised by the resulting images (which can also be viewed on YouTube), because they revealed complex networks of interconnected PanINs even in tissue considered healthy, with an average overall burden of 13 PanINs per cubic centimetre. In essence, this study shows that the normal adult pancreas harbours hundreds of PanINs, or at least appears to, almost all of which, as verified by the researchers, with mutations of the KRAS (Kirsten rat sarcoma viral oncogene homolog) gene, which plays a decisive role in certain forms of cancer, including pancreatic ductal adenocarcinoma.
“Not many people actually develop pancreatic cancer, so we were shocked to find a lot of precancer, or PanINs, within the normal regions of the pancreas,” said Laura Wood, associate professor of pathology and oncology at Johns Hopkins University and a senior author of the study. “This research highlights what we don't yet know about normal ageing and raises fundamental questions about how cancer arises in the human pancreas.”
A huge variety of mutations
The researchers investigated eight samples in greater detail via microdissection guided by the 3D reconstruction, and by analysing the DNA sequences of certain PanINs. This analysis showed that the precancerous lesions are genetically distinct from one another (driven by different mutations), even if some genes, such as the altered KRAS gene, are present in almost all of the samples, as mentioned earlier. Until now, in other organs, precancerous lesions driven by such different mutations had practically never been encountered. This clearly complicates matters, making it more difficult to identify unique markers for the tumour.
The significance of this discovery, and above all the presence of so many PanINs in healthy tissue, is yet to be understood, but there are now solid foundations in place for future research.
What is more, CODA (a tool that automatically aligns images of the pancreas) does not only work on the pancreas, and could therefore also improve the diagnosis of cancers of other organs.
“This is just the beginning,” confirmed Laura Wood. “If normal tissue has thousands of PanINs, then how do we identify which ones are clinically relevant to disease and which ones are not? Better understanding the early precursors to cancer through detailed and anatomic molecular maps is the first step. Until you look in 3D, you don’t know what you’re missing.”
