Can using radiomics during screening mammography improve breast cancer diagnosis?

By using radiomics, Chinese researchers found that the diagnostic performance of mammography could improve and offer complementary information to radiologists regarding benign and malignant breast tumors, as reported in the Journal of the American College of Radiology on Dec. 5.  

Quantitative image analyses, such as computer-aided diagnosis (CAD) algorithms, are continually being developed to help radiologists more objectively and clearly interpret mammograms.  

Radiomics, an extension of CAD, has proven to capture texture features that describe the intensity distributions within lesions, spatial and spectral frequency patterns and can help characterize the relationships among different intensity levels within the lesion. But these features are not always visible to a radiologist’s naked eye, especially in mammograms.  

“The moderate specificity of mammography might result in additional imaging examinations, costs and patient anxiety and may cause avoidable breast biopsies,” Nan Hong, MD, PhD, radiologist at Peking University People’s Hospital in Beijing, and colleagues. “Therefore, CAD models might be used as an adjunct to the radiologists’ assessments.”  

The researchers included 173 patients (74 with benign lesions and 99 with malignant lesions) who underwent screening mammography before neoadjuvant chemotherapy treatment. Using pre-processing methods including centering and normalization, the researchers extracted radiomic features from each patient’s mammogram.  

Additionally, four artificial intelligence (AI)-based machine learning algorithms—called support vector machine, logistic regression, K-nearest neighbor and Bayes classification—were trained with a separate dataset and used to create a predictive model. The classification performance was then compared with the predictions of two breast radiologists, according to the researchers. A total of 51 radiomic features remained after pre-processing.  

Study results included the following:  

  • The diagnostic accuracy, specificity and sensitivity of the logistic regression model for the training data set were 0.978, 0.975, and 0.983, respectively.
  • The diagnostic accuracy, specificity and sensitivity for the testing data set were 0.886, 0.900, and 0.867, respectively.
  • The accuracy, specificity and sensitivity of the combined reading of the two radiologists were 0.772, 0.710, 0.862 respectively in the training data set, and 0.769, 0.695, 0.858, respectively in the testing data set.  

Of the four regression models, logistic regression classification performed with the greatest accuracy, specificity and sensitivity. Overall, the imaging technique was standardized and uniform across the study population, according to the researchers.