Cancer screenings and testing are really determined by data. Reams of data. Unbelievably large mountains of data. And few technologies can sort through such substantial quantities of data like artificial intelligence.
Microsoft is partnering with a pathology company, Paige, to create a massive image-based AI model focused on identifying different types of cancer. The model is already being training on more than 4 million images. With (even more) data and time, these models could be used to screen more patients more quickly than ever before, ultimately reducing costs. The cost reduction may allow for earlier or more frequent testing.
There are already positive results. One clinical trial used AI to identify instances of breast cancer during routine screenings. The trial compared the test of two groups:
- One group involved a standard reading by two professional radiologists.
- The second group involved 1 radiologist and 1 AI model.
The AI-assisted team ultimately reduced total workload by 44% – and detected 20% more tumors.
The Drawbacks of AI in Cancer Screenings
More accurate and, one day, more affordable cancer screening will help identify deadly cancers earlier, increasing the odds of successful treat. But the improved screening capacity of AI could lead to an unanticipated problem – over-diagnosis.
Read more: Are US Cancer Death Rates Declining?
Increased access to screenings can, in some cases, cause health issues, such as injury during colonoscopies or infections after biopsies. And not all tumors are lethal, which may increase patient anxiety unnecessarily. As a result, too many diagnosis may contribute to negative outcomes for patients who might otherwise be unaffected by a tumor.
The solution? Adjusting how we teach AI to spot potential tumors or cell structures in the first place. Instead of a binary, yes vs. no identification, researchers and AI engineers will need to focus on the algorithm on specific types of cancer, cancer structures, or cancers in particular locations that pose great risks.
Researchers are already experimenting with ways to generate more useful data from AI – and humans will be intimately involved with the cancer screening process for many, many years to come.