A recent study shows that AI support significantly enhances radiologists’ accuracy in detecting breast cancer during mammogram screenings, underscoring the potential of AI technology to transform medical diagnostics.
A recent study published today in the journal Radiology reveals that artificial intelligence significantly improves the accuracy of radiologists in detecting breast cancer from screening mammograms. This boost in diagnostic performance allows radiologists to focus more keenly on suspicious areas, potentially saving lives through earlier cancer detection.
Previous studies have indicated that AI can assist radiologists by increasing cancer detection sensitivity without lengthening reading times. However, the impact of AI on radiologists’ visual search behaviors had not been thoroughly studied until now.
To investigate this, the research team utilized an eye-tracking system to analyze how radiologists’ reading patterns changed with the support of AI. The setup included a camera-based device with infrared lights to track eye movements, thus pinpointing the exact locations on the mammograms where radiologists were focusing.
“By analyzing this data, we can determine which parts of the mammograms the radiologists focus on, and for how long, providing valuable insights into their reading patterns,” joint first author Jessie J. J. Gommers, from the Department of Medical Imaging at Radboud University Medical Center in Nijmegen, Netherlands, said in a news release.
In the study, 12 radiologists examined mammography results from 150 women, half of whom had breast cancer.
The results were compelling: radiologists demonstrated higher breast cancer detection accuracy with the aid of AI compared to readings without AI assistance. There was no observed difference in the mean sensitivity, specificity or reading time thanks to the enhanced precision provided by AI.
“The results are encouraging,” Gommers added. “With the availability of the AI information, the radiologists performed significantly better.”
The study found that when using AI support, radiologists spent more time examining regions that contained actual lesions. The AI system helped radiologists streamline their focus: low AI scores allowed them to swiftly dismiss non-cancerous cases, while high scores prompted a more meticulous investigation of suspicious regions.
“Radiologists seemed to adjust their reading behavior based on the AI’s level of suspicion: when the AI gave a low score, it likely reassured radiologists, helping them move more quickly through clearly normal cases,” added Gommers. “Conversely, high AI scores prompted radiologists to take a second, more careful look, particularly in more challenging or subtle cases.”
The AI’s markings served as visual cues, effectively acting as an additional set of eyes and guiding radiologists to potentially worrisome areas.
“Overall, AI not only helped radiologists focus on the right cases but also directed their attention to the most relevant regions within those cases, suggesting a meaningful role for AI in improving both performance and efficiency in breast cancer screening,” Gommers added.
While AI’s contributions are promising, Gommers warned of the potential dangers of over-relying on AI, which could lead to missed cancers or unnecessary additional imaging. The key lies in AI accuracy and proper radiologist training.
“Educating radiologists on how to critically interpret the AI information is key,” she emphasized.
Looking ahead, the researchers are conducting more studies to determine the optimal timing for providing AI information and exploring methods to predict AI uncertainty. This would allow AI support to be used more selectively, enhancing its value in clinical settings.
“This would enable more selective use of AI support, applying it only when it is likely to provide meaningful benefit,” added Gommers.
This research showcased a collaborative effort of experts from Radboud University and other institutions, promising a future where AI can significantly augment medical diagnostics, particularly in the critical field of breast cancer screening.