Skip to main content

22-01-2012 | Gynaecology | Article

CAD system identifies malignant breast lesions on MRI


Free abstract

MedWire News: Magnetic resonance imaging (MRI) research shows that a computer-assisted diagnosis (CAD) system can reliably differentiate between malignant and benign breast tissue.

"Observer-independent CAD may be a promising additional tool for the interpretation of breast MRI in the clinical routine," say Diane Renz (Charité University Medicine Berlin, Germany) and co-authors.

The researchers compared a novel automatic CAD technique to analyze MRI dynamic and morphologic markers for breast cancer in 108 patients, aged an average of 51.3 years old, with 88 histologically confirmed malignant tumors and 53 benign tumors.

The CAD system gave each lesion a probability of malignancy, with a morpho-dynamic index (MDI) of between 0 and 100%, after processing nonrigid motion correction, detection of lesions using a segmentation algorithm, and dynamic and morphologic parameters.

As reported in the Journal of Magnetic Resonance Imaging, the CAD system was accurate for the prediction of malignant tumors.

The CAD system incorrectly identified two invasive multifocal carcinomas as benign, and incorrectly labeled three fibroadenomas and one atypical ductal hyperplasia as malignant.

The correctly identified malignant tumors had a significantly higher average MDI than the correctly identified benign tumors, at 86.1% and 41.8%, respectively.

Using an MDI cutoff for malignancy of 50%, the CAD system was 96.5% sensitive and 75.5% specific for the identification of malignant tumors using MRI, with a receiver operator curve diagnostic accuracy of 93.5%.

"The goal of this study was not to replace the radiologist but rather to offer a supplementary diagnostic tool for detecting and classifying tumors," explain Renz et al.

" The consideration of morphologic CAD features may be particularly helpful for identifying carcinomas lacking typical malignant kinetic patterns."

They conclude: "Further investigations should compare the influence on the diagnostic performance of human readers without and afterwards with the assistance of the fully automatic technique and should also include the evaluation of nonmass-like lesions."

By Lynda Williams

Related topics