Purpose: CAD remains a useful adjunct tool for the interpretation of DCE MRI of the breast. During lactation period however, DCE is hampered by the prominent physiological background parenchymal enhancement (BPE) which reduces its accuracy for breast cancer detection and diagnosis. Our aim was to investigate the current performance of CAD among the challenging cohort of lactating patients, and accordingly to establish the framework for future development of a more reliable means of lactation-oriented CAD.
Materials and Methods: The efficacy of commercial CAD software (GE) was retrospectively tested using datasets DCE MRI of lactating patients diagnosed with breast cancer (n=34) and a matched control group of healthy lactating patients. In addition, further analysis of DCE was performed for selected cases, in which region of interest of the biopsy-confirmed lesion was manually drawn on the first subtraction image (t1-t0) and was automatically transferred to the following four post injection time points, enabling the quantitative evaluation of the dynamic enhancement patterns in the tumor, as well as in the surrounding normal lactating parenchyma.
Results: Initial observation showed that while the conventional CAD software did detect most breast cancer lesions, it also frequently flagged additional normal BPE regions in the cancer well as in the healthy control datasets, overall resulting in poor specificity. Furthermore, preliminary analysis of the quantitative enhancement dynamics suggested possible differences in patterns of wash-in and wash-out of DCE between the normal lactating BPE and breast cancer. Thereafter, an algorithm for lactation-sensitive CAD was attempted, taking into account the DCE characteristics of the two environments and omitting noisy clusters.
Conclusion: Current CAD definitions of DCE MRI of are not useful the diagnostic work-up of the lactating breast. New procedure for achieving high-definition lactation-tailored CAD was developed and demonstrated.
