Researchers have developed a new artificial intelligence (AI) model that could help make radiologists more accurate when evaluating ultrasound images for signs of breast cancer. The group shared its results in Radiology: Artificial Intelligence, noting that the algorithm could be especially helpful for novice readers who are still relatively inexperienced.[1]
Conducted across four hospitals, the retrospective study dataset included data from more than 45,000 ultrasound images acquired using 42 different types of machines.
The group’s deep learning (DL) model, a dual attentin-based convolutional neural network, model was challenged to distinguish between malignant and benign breast. As a comparison, three novice readers with less than five years of experience with ultrasound imaging and two experienced readers with 8 and 18 years of experience, respectively, interpreted 1,024 randomly selected images of lesions.
Overall, the DL model achieved an expert-level diagnostic performance, showcasing an area under the receiver operating characteristic curve (AUC) of 0.94 for the internal dataset. The AUCs for external datasets were 0.92, 0.91, and 0.96 for three different hospitals, indicating the model's consistent performance across diverse settings.
Key findings from the team’s analysis included:
As experienced as an expert: The DL model demonstrated similar performance to experienced human readers, indicating its potential as an accurate diagnostic tool. Specifically, the DL model's AUC was comparable to that of experienced radiologists.
Aided novice readers: Novice radiologists with less than five years of ultrasound experience showed substantial improvements when assisted by the DL model. The model increased their diagnostic accuracy, effectively helping them achieve performance levels similar to those of experienced readers.
Reduced false-positive rates: While ultrasound is commonly used for breast cancer diagnosis due to its availability and cost-effectiveness, its accuracy remains a challenge, often resulting in high false-positive rates and unnecessary biopsies. With the DL model's assistance, diagnostic accuracy and interobserver agreement experienced significant improvements among both novice and experienced radiologists. Particularly noteworthy was the 7.6% reduction in the average false-positive rate.
In summary, the study's findings indicate that DL-assisted diagnosis may be highly beneficial in the field of breast tumor diagnosis using ultrasound images. The model's accuracy, consistent performance across different hospitals, and ability to aid both novice and experienced readers suggest a promising future for integrating DL technology into clinical practice. By improving diagnostic accuracy and reducing false-positive rates, the authors wrote, the DL model holds potential to streamline clinical workflows and reduce the risk of performing unnecessary biopsies.
“This method is promising as an efficient and cost-effective tool for assisting radiologists, especially novice radiologists, in breast tumor diagnosis,” wrote first author Huiling Xiang, a specialist with the department of ultrasound at the Collaborative Innovation Center for Cancer Medicine in China, and colleagues. “Further studies are warranted to characterize the feasibility of the model's widespread adoption.”
Read the full study here.
Chad is an award-winning writer and editor with over 15 years of experience working in media. He has a decade-long professional background in healthcare, working as a writer and in public relations.