AI Saves Critical Time in Cancer Detection
For years, artificial intelligence pioneers have predicted that AI-powered tools would transform the practice of medicine. While we are nowhere close to a full transformation, these tools are inching closer to routine clinical use. An important step in that path is comparing the performance of AI-based tools and medical experts in detecting health problems.
Interested in exploring the risks and rewards of the AI revolution? Join us in Lake Nona, FL at Techonomy 23: The Promise and Peril of AI.
Recently, several new studies have been conducted to find out what happens when you pit AI against skilled medical professionals. The results show remarkable potential for augmenting clinicians’ ability to diagnose illness efficiently.
These studies focus on cancer, which is often detected by spotting irregularities on imaging scans. That kind of pattern recognition is squarely in the wheelhouse of AI, making it a good initial application for understanding how these tools might improve healthcare.
Two large studies in Sweden examined the outcomes of mammography-based breast cancer detection. One included more than 80,000 women, and the other nearly 56,000 women. In both cases, mammogram results for a control group were analyzed by two radiologists (the standard in Sweden), while results for the AI test group were analyzed by one radiologist plus an AI detection tool. The smaller study also tested the AI analysis on its own, without any human radiologist involved, as well as readings by two radiologists plus AI.
While AI champions might hope that the computational tool would trounce human performance, that’s not what happened. Instead, both studies found that augmenting the radiologist with an AI tool improved detection by a relatively small amount. The larger study reported that replacing one of the two traditional radiologists with AI led to similar cancer detection rates as having two radiologists, while the smaller study cited a 4% increase in cancer detection with the implementation of AI. When used alone, AI detection also led to a decrease in false positives, which in regular practice could reduce the number of unnecessary biopsies for patients.
But in both cases, the real value of the AI tool was in saving time. The studies reported that the groups evaluated with AI and a radiologist had reduced workloads of 44% to 50%. If that result held up in routine clinical practice, a single radiologist would be able to screen nearly twice as many patients for cancer compared to standard practices. As more hospitals and healthcare systems are affected by a global shortage of radiologists—particularly as increased cancer screenings mean higher demand for these specialists—any tool that improves efficiency without compromising patient care could prove advantageous.
Additional trials will be needed before AI tools can be implemented in regular practice for analyzing mammograms, but these studies strongly suggest that AI will have an important role to play in improving healthcare through routine cancer screenings.