Breast MRI Computer-Aided Diagnosis Systems

Abstract

Breast MRI, or BMRI, has emerged as a promising technique in detecting, diagnosing, and staging breast cancer. It has shown a high sensitivity and moderate specificity in detecting breast cancer. Mammographic screening is the current method of choice for early breast cancer detection. However, it has been shown that standard mammography misses 10% to 30% of cancers that are visible using breast MR. The superior detection sensitivity of BMRI has been attributed to its tomographic properties, greater soft-tissue contrast, and larger field of view. This enables BMRI to be particularly useful for tumors located high on the chest wall or low in the axillary tail of the breast. In addition to its improved detection capabilities, BMRI gives a better description of the exact size and extent of a tumor than mammography or palpation. With all of these benefits, you might expect to see radiology rushing to adopt BMRI. The excitement is tempered, though, because BMRI has higher costs, limited availability, and generates large data sets that can be difficult to analyze. Also, MR is a qualitative imaging modality. This means that in repeat scans of the same subject, relative contrast differences between regions are preserved, but absolute gray-level measurements may differ. All these together make it unlikely that BMRI will replace mammography entirely, but BMRI could become the screening modality of choice for high-risk patients. So why is there a need for computer-aided diagnosis (CAD) system to assist radiologists? As touched on briefly before, BMRI can be difficult to analyze. In 1994, Vyborny and Giger showed that if two radiologists read the same mammogram, or even if the same radiologist rereads a mammogram, diagnostic accuracy improves. A CAD system can be a second set of “electronic” eyes. If properly designed, it will also overcome the difficulties mammographers face when presented with the unfamiliar territory that is BMRI. At present, BMRI is best used as a diagnostic modality for high-risk patients. As technology improves, cost generally decreases, and advanced CAD systems can reduce the radiologist's workload. CAD systems have demonstrated high specificity for detected lesions. This enables BMRI to be a noninvasive alternative for handling suspicious lesions detected by mammography and could improve an existing technology (MRI) rather than require new technologies to be developed. Mammography CAD systems have been used for: detecting breast masses, finding microcalcifications, and classifying tumors. There have been many different methods for classification, including: wavelets, fractals, statistical methods, vision-based methods, and artificial neural networks. Overall, there are not nearly as many publications on BMRI CAD systems as for mammography. For more information about MRI principles, readers may look at Ref. 27, and for more information about breast MRI fundamentals and protocols, readers are referred to Refs. 28 and 29.