Hi guys, it’s DataScienceKid here. Today, we are going to take a look at how AI is revolutionising a very sad domain, which is pancreatic cancer. Pancreatic cancer is one of the deadliest cancers out there. In 2020 alone, around 495 000 people were diagnosed with pancreatic cancer. On top of that, unfortunately, only around 5% of those diagnosed with this peculiar cancer have a chance of surviving. In the chart below, sadly only 9% of patients have a chance of surviving 5 years after being diagnosed!
On top of being extremely difficult to find and diagnose, it is also hard to treat. At the stage that pancreatic cancer is most easily treated, there are usually no noticeable symptoms, and is very hard to detect. But that’s where AI comes in.
Recently, researchers at Nature Medicine were able to find a breathtaking discovery. Researchers were able to train an AI model to identify patients at risk around three years before their diagnosis, solely by using their medical records! This could completely level the playing field, and prevent thousands of lives from being taken by this deadly cancer.
So, how exactly did they make this work? Well, the first step was to train the AI on around 6.2 million health records! These records came from the Denmark National Health system, and spanned back 41 years. Of these 6 million patients, around 24 000 of them contracted pancreatic cancer. From this dataset, the AI was able to spot specific patterns from past diseases and certain medical conditions. For example, things such as gallstones, anemia, and diabetes can often be signs of pancreatic cancer.
The researchers advise against assuming that any one of these diagnoses is predictive or responsible for developing pancreatic cancer in the future. However, the pattern and order in which these things appear over time provide cues for an AI-based surveillance model, which may lead doctors to closely watch patients at high risk or order tests accordingly.
After having trained the Ai with this dataset from Denmark, researchers decided to try it on a dataset from the U.S. Veterans Health Administration. The Ai was way less accurate on this dataset, but after retraining the Ai with this american data, the accuracy improved marginally.
So, now you know a little bit about this Ai, and the amazing things that it might be able to do and the lives it could potentially save. These kinds of discoveries are exactly why Ai is so powerful and is the tool of the future. Tune in next time for more interesting articles like this one!