An AI tool that improves diagnosis of ultrasound breast images could reduce unnecessary biopsies and patient anxiety
"If our efforts to use machine learning as a triaging tool for ultrasound studies prove successful, ultrasound could become a more effective tool in breast cancer screening.” Dr Linda Moy, professor of radiology at NYU Grossman School of Medicine
An artificial intelligence (AI) algorithm has been successful in reducing the number of false positives in breast ultrasound, which could lead to one in four biopsies being avoided.
The study, carried out by researchers at New York University (NYU), and published in Nature Communications, involved training a deep-learning algorithm on nearly 290,000 breast ultrasound images. The images used were taken from 143,203 patients examined between 2012 and 2018 at NYU Langone Health.
After training the AI tool on the data, researchers tested it on 44,755 already completed ultrasound examinations. They then compared the performance of the tool with that of 10 radiologists, who each reviewed a separate set of 663 breast exams from the internal test set, achieving an average accuracy in diagnosing breast cancer of 92%. When aided by the AI model, accuracy increased to 96%. All diagnoses were checked against tissue biopsy results.
The researchers then created a simulated hybrid prediction model that combined the predictions of the radiologists and the AI tool. Applying this to the images, they found that the model would have decreased the average radiologist’s false-positive rate by 37.3%, and the number of biopsies needed to confirm suspected tumours by 27% – while maintaining the same level of sensitivity.
Ultrasound plays an important role in breast cancer diagnosis. Unlike mammography, it does not involve exposure to radiation, and is better for penetrating dense breast tissue and distinguishing packed but healthy cells from compact tumours.
Interpretation of the images is challenging for radiologists, however, leading to a good deal of variation and a high proportion of false positives. In fact, some studies have shown that a majority of breast ultrasound examinations indicating signs of cancer turn out to be non-cancerous after biopsy.
For women, false breast cancer diagnoses can create a good deal of anxiety and involve unnecessary additional procedures. If the initial results are borne out, the AI tool would reduce much of the anxiety and the need for biopsies.
“If our efforts to use machine learning as a triaging tool for ultrasound studies prove successful, ultrasound could become a more effective tool in breast cancer screening, especially as an alternative to mammography, and for those with dense breast tissue,” said Dr Linda Moy, a co-investigator on the study and professor of radiology at NYU Grossman School of Medicine. “Its future impact on improving women’s breast health could be profound.”
Dr Krzysztof Geras, the senior investigator on the study, said that, while his team’s initial results were promising, clinical trials of the tool in current patients and real-world conditions are needed before it can be routinely deployed. He also plans to refine the AI software to include additional patient information, such as a woman’s added risk from having a family history or genetic mutation tied to breast cancer.