AI Matches Radiologist Performance on Mammograms and DBT Exams: A Meta-Analysis
The ever-increasing rate of breast cancer cases around the world has made mammography an essential screening tool for identifying the early stages of the disease. However, the accuracy of mammograms can vary from one radiologist to another, and the interpretation of images can be tedious and subjective. This has led to the development of artificial intelligence (AI) that can accurately and quickly interpret mammograms. We will delve into a recent meta-analysis that determined the performance of AI versus radiologists in breast imaging exams.
The research, published on May 23 in the journal Radiology, compared over 1.1 million screening breast imaging exams interpreted by AI versus radiologists. The findings of the meta-analysis revealed that AI performs equally or better than radiologists in screening mammograms and digital breast tomosynthesis (DBT) exams. Specifically, the AI system demonstrated a slightly higher pooled area under the curve (AUC) than individual radiologists when it came to mammograms and DBT exams.
The team led by Dr. Jung Hyun Yoon, PhD, from Yonsei University in Seoul, highlighted that critical evaluations are needed before AI can be integrated into routine breast cancer screening. The researchers also pointed out that AI’s performance has to reach a certain level before it can enhance screening outcomes and reduce the workload of radiologists. Besides, more comparative DBT studies are essential to determine the accuracy of AI in detecting breast cancer.
According to the meta-analysis, standalone AI for screening digital mammography performed at least as well or even better than individual breast radiologists on average reader outcomes. The projected role of AI software in interpreting mammograms is to reduce the reader variability, increase the number of false-negative cases, and reduce the rate of unnecessary biopsies.
Several AI software tools have been developed and validated for use in breast imaging, such as Transpara, iCAD Second Look, and Quantra. Still, the researchers emphasized that the technology’s efficacy should be critically evaluated before standalone implementation. Despite the progress made in the field of AI applied to mammography, the ultimate goal remains the same: to improve the detection of breast cancer at an early stage.
The comparison of the performance of AI versus radiologists in breast imaging exams indicates the promise of AI in significantly reducing the workload of radiologists while improving screening accuracy. The meta-analysis published in Radiology showed that AI performs equally or even better than radiologists in screening mammograms and DBT exams. While AI technology holds much promise in the diagnosis of breast cancer, researchers still urge caution and critical evaluation before standalone implementation. Further DBT studies testing the efficacy of AI in breast cancer detection will ascertain AI technology’s role in breast imaging and help improve breast cancer diagnosis and treatment.