Optimizing CAD to improve early detection in high-risk breast cancer patients

Twitter icon
Facebook icon
LinkedIn icon
e-mail icon
Google icon
 - Breast screening

Using a two-stage classifier to differentiate between malignant or benign mass and non-mass lesions can improve the accuracy of computer-aided diagnosis (CAD) systems and help identify cancer earlier in high-risk patients, according to results of a study published in the March issue of the journal Radiology.

The key to improving breast cancer outcomes is to boost the early-detection capabilities of existing breast imaging modalities. One way of achieving this goal is through utilization of CAD systems to evaluate the results of contrast-enhanced MRI and to analyze the nature of lesions found therein.

But like CAD systems, not all lesions are created equal, said lead author Cristina Gallego-Ortiz, MSc, of the University of Toronto, and Anne Martel, PhD, of the Sunnybrook Research Institute in Toronto.

“According to the Breast Imaging Reporting and Data System (BI-RADS) lexicon, enhancement patterns can be classified as mass, nonmass, or foci (,5 mm) enhancing lesions,” they wrote. “Because the physiologic basis of enhancement between mass and nonmass lesions is highly variable, when designing a CAD for breast MR imaging screening, it may be more effective to separately optimize classifiers for mass and nonmass lesions to improve the accuracy of CAD.”

Gallego-Ortiz and Martel set out to explore this hypothesis, conducting a study aimed at determining an optimal classifier design for both types of lesions in a CAD system to be used in conjunction with contrast-enhanced MRI. To do so, they performed a retrospective analysis of imaging studies from 280 histologically proved mass lesions and 129 histologically proved non-mass lesions using BI-RADS classifications and the existing image data. They then calculated which classifiers performed best at predicting whether the lesions were benign or malignant.

Their results showed that in the 176 features extracted from the images, the best predictive performance belonged to a two-stage cascade classifier using mass versus non-mass classification followed by an assessment malignancy. This design improved upon the diagnostic results of the one-shot classifier (all benign versus malignant classification) and reduced misclassifications by 12 percent.

“Separately optimizing feature selection and training classifiers for mass and non-mass lesions improves the accuracy of CAD for breast MR imaging,” the authors concluded. “Our cascaded classifier may provide an advantage for screening women at high risk for breast cancer, in whom the ability to diagnose cancers at an early stage is of primary importance.”