Classification and calculation of breast fibroglandular tissue volume on SPGR fat suppressed MRI

This paper presents an automatic method to classify and quantify breast fibroglandular tissues on T1 weighted spoiled gradient-echo (SPGR) fat suppressed MRI. The breast region is segmented from the image using mathematical morphology, region growing, and active contour models. The breast-air and breast-chest wall boundaries are located using smooth and continuous curves. Three tissue types are defined: fatty tissue, fibroglandular tissue, and skin. We then employ a fuzzy C-means (FCM) method for tissue classification. For each pixel inside the breast region, the normalized pixel intensity and normalized distance to the breast-air boundary are computed. These two values form a two-dimensional feature space. A fuzzy class is defined for each tissue type. The initial centroid for each class is obtained from training images. The pixel membership values indicate the possibility of a pixel belonging to a certain tissue class. Pixels with highest membership in the fibroglandular class are then classified as fibroglandular tissue. We have tested our method on 29 patients. We automatically segmented the breasts and computed the volume percentage of fibroglandular tissue for both left and right breasts. We then compared the calculated tissue classification with manually generated tissue classification by two experienced radiologists. The two results agreed on 94.95% of breast segmentation, and the average fibroglandular percentage difference is about 3%. This method is useful in research studies assessing breast cancer risk.