Google has now published a new paper detailing the progress that has made in improving the accuracy and usefulness of its cancer detecting AI algorithms. Specifically, the new study shows that artificial intelligence being developed by the company can be used to help expand on the number of lung cancer screenings worldwide and save lives by getting the fight started earlier.
The search giant notes that as many as 160,000 deaths were the result of lung cancer in the US in 2018 and that screening using low-dose computed tomography can reduce mortality by between 20-percent and 43-percent. Reading those scans is where the company’s AI enters the equation. In as many as 6,716 National Lung Cancer Screening Trial cases and in clinical validation of 1,139 cases, the company says its latest model performs at 94.4-percent “area under the curve.”
In layman’s terms, that equates to an outperformance of the radiologists involved in the study, resulting in an 11-percent reduction to false positives and a 5-percent reduction in false negatives.
Providing the assist
The research in question started back in 2017 and is one of many cancer-related AI research projects started over the past several years by Google.
The tools are not meant to entirely replace radiologists — at least not yet — but may be able to serve as a powerful tool for catching cancer earlier and starting treatment sooner to save countless lives. In this case, the company compared its artificial intelligence program against radiologists across more than 45,856 de-identified chest CT screening cases. That gives some indication of the problem facing the specialists and where AI could help.
A radiologist will generally need to look through hundreds of those types of images for a single scan, presented as 2D slices of a patient’s internal anatomy. In this case, that’s the lungs. The discrepancies that are being sought in those images are often tiny and only show up a small percentage of the images, making it easier to catch the growth of malignant tissue and cells as cancer progresses.
Not only can an AI assistant scan through those types of images looking for problems, taking some of the load off of radiologists. Google says its technology can easily make comparisons to previous scans, tracking changes over time and it can utilize 3D volumetric modeling to get a better overall picture of those changes.
The AI compares findings to existing datasets, helping it to more easily identify problems and predict the likeliness that a given node or anomaly will become or is already malignant. The results of that are the figures shown in the new study, where radiologists assisted by the AI perform significantly better compared to those without that assistance.
What does this really mean?
The advancement shows that there may be a place for AI to help in increasing the number of scans that take place yearly, due to the plausible reduction in terms of radiologists’ workloads. As things currently stand, Google says that only around two to four percent of ‘eligible’ patients are scanned for lung cancer. So the technology could potentially save tens of thousands of lives in the US alone.
That doesn’t mean AI-assisted cancer screenings are going to become widespread anytime soon. The publishing of a research paper is only one of the first steps in the process involved in bringing the technology to the masses and the work will need to undergo rigorous peer review and further testing before it’s ready for prime time.