Fully automated breast segmentation on spiral breast computed tomography images

Abstract

Introduction: The quantification of the amount of the glandular tissue and breast density is important to assess breast cancer risk. Novel photon-counting breast computed tomography (CT) technology has the potential to quantify them. For accurate analysis, a dedicated method to segment the breast components-the adipose and glandular tissue, skin, pectoralis muscle, skinfold section, rib, and implant-is required. We propose a fully automated breast segmentation method for breast CT images.

Methods: The framework consists of four parts: (1) investigate, (2) segment the components excluding adipose and glandular tissue, (3) assess the breast density, and (4) iteratively segment the glandular tissue according to the estimated density. For the method, adapted seeded watershed and region growing algorithm were dedicatedly developed for the breast CT images and optimized on 68 breast images. The segmentation performance was qualitatively (five-point Likert scale) and quantitatively (Dice similarity coefficient [DSC] and difference coefficient [DC]) demonstrated according to human reading by experienced radiologists.

Results: The performance evaluation on each component and overall segmentation for 17 breast CT images resulted in DSCs ranging 0.90-0.97 and in DCs 0.01-0.08. The readers rated 4.5-4.8 (5 highest score) with an excellent inter-reader agreement. The breast density varied by 3.7%-7.1% when including mis-segmented muscle or skin.

Conclusion: The automatic segmentation results coincided with the human expert's reading. The accurate segmentation is important to avoid the significant bias in breast density analysis. Our method enables accurate quantification of the breast density and amount of the glandular tissue that is directly related to breast cancer risk.

Keywords: CT; breast; density; segmentation.

FIGURE 1

Eight exemplary spiral breast computed tomography (CT) images in coronal (top) and transverse or sagittal (bottom) planes

FIGURE 2

Breast component segmentation by readers (top) and our automatic segmentation algorithm (bottom) for the pectoralis muscle and skin (a) and (b) and skinfold section (c)

FIGURE 3

Schematic Hounsfield unit (HU) distribution of the adipose and glandular tissue in a real breast computed tomography (CT) image (a) and box‐and‐whisker diagram for the HU distribution of the breast components (b). In (b), the boxes represent the IQR, the red lines the median, and the blue error bar 95% HU range.

FIGURE 4

Framework of the proposed segmentation method

FIGURE 5

Demonstration of segmentation algorithm's process to investigate the seed region and segment the component accordingly (from top to bottom) for the skinfold section (a), silicone and rib (b), pectoralis muscle (c), skin (d), and glandular tissue (e). First row: (a), (c), and (d)—red contour represents the seed region; (b)—green contour the region to investigate the seed; (e)—green contour the confined region to segment the glandular tissue. Second row: (a), (c), and (d)—red and blue contours represent the markers to apply the adaptive watershed algorithm; (b)—red and blue contour the seeds of the respective objects; (e)—red paint the seed region. Third row: (a), (c), and (d)—red contour represents the final segmentation of the object; (b)—red and blue contours the final segmentation of the respective objects; (e)—red and blue paint together the final segmentation.

FIGURE 6

Segmented breast computed tomography (CT) images corresponding to the respective examplary images [a‐h] in Figure 1

FIGURE 7

Subjective evaluation result in five‐point Likert scale

FIGURE 8

Glandularity plot (a) and box‐and‐whisker diagram (b) of the estimated glandularity differences compared to the correct Hounsfield unit (HU)‐derived estimation and the volumetric segmentation‐based estimation or the erroneous HU‐derived estimation with the mis‐segmented skin, pectoralis muscle, or skinfold section.