Who Better Detects Breast Cancer on Ultrasound Exams: AI or Radiologists? Both.

In a new study on ultrasound screening for breast cancer, a team of machine learning researchers, imaging scientists, and radiologists at NYU Grossman School of Medicine, NYU Center for Data Science, and NYU Abu Dhabi—including investigators from the Center for Advanced Imaging Innovation and Research—have found that a “hybrid” of radiologists and a deep learning algorithm has yielded a lower biopsy rate and higher accuracy than that achieved by either the human experts or the artificial intelligence (AI) algorithm alone.

The findings, which at the time of this writing await peer review, echo the results of a 2019 study of mammography exams and strengthen the case that machine learning has the potential to help radiologists more accurately diagnose the most common malignancy among women. The 2019 research—conducted by investigators at NYU Grossman School of Medicine, NYU Center for Data Science, the Center for Advanced Imaging Innovation and Research, and colleagues at several other departments—showed that mammography cancer screening evaluations conducted by radiologists and an AI model, which rivaled each other in accuracy, became more precise when combined than they were for either the radiologists or the AI alone.

But before collaborations between medically trained human experts and AI can benefit patients, the technology has to become more transparent.

To learn more about the new study and how researchers are trying to bring algorithms and people together, open the visual story below.

Visual Story

AI Can Reduce Overdiagnosis in Ultrasound Screening for Breast Cancer

A simulated collaboration between radiologists and an artificial intelligence algorithm leads to fewer calls for unnecessary biopsies and boosts accuracy in identifying cancer on ultrasound images.

Update
September 24, 2021

Research featured in this post has been peer reviewed and published.