Google's AI for lung cancer diagnosis proves more accurate than radiologists in early trial
Like all forms of the disease, an early diagnosis of lung cancer can greatly improve a patient's chance of survival, but like all cancers, this is much easier said than done. A Google research initiative aimed at harnessing artificial intelligence to better model and predict lung cancer has shown promise in a newly published study, with the technology even outperforming certified radiologists in some regards.
Breast cancers, skin cancers and ovarian cancers, are just a few types of cancer that could be better treated through the assistance of AI, with a string of research breakthroughs in recent years raising the prospect of earlier and more accurate diagnosis.
By training machine learning algorithms on thousands of medical images, these systems can detect small and potentially problematic changes that might go unnoticed by humans. And we are starting to see how the technology could one day lead to far better patient outcomes.
In Google's case, it trained its machine learning algorithm on more than 45,000 chest CT scans, taken from the National Health Institute and Northwestern University, some of which featured cancer in various stages. The algorithm works by generating a 3D model from the CT scan, using that to detect tiny malignant tissue in the lung nodules that would otherwise be difficult to spot and produce an overall lung cancer malignancy prediction.
The algorithm was then put to work using a single CT scan for diagnosis, with the accuracy of the algorithm compared to that of six board-certified radiologists. Google says it detected five percent more cancer cases and reduced false-positives by more than 11 percent.
Though the system has only been validated on existing patients using historical scans and further work is needed to explore how it will function in a clinical setting, Google describes the early results as "encouraging."
"This creates an opportunity to optimize the screening process via computer assistance and automation," the authors write in their paper abstract. "While the vast majority of patients remain unscreened, we show the potential for deep learning models to increase the accuracy, consistency and adoption of lung cancer screening worldwide.'
The research was published in the journal Nature Medicine.