Strain histograms are equal to strain ratios in predicting malignancy in breast tumours

Abstract

Objectives: To assess whether strain histograms are equal to strain ratios in predicting breast tumour malignancy and to see if either could be used to upgrade Breast Imaging Reporting and Data System (BI-RADS) 3 tumours for immediate biopsy.

Methods: Ninety-nine breast tumours were examined using B-mode BI-RADS scorings and strain elastography. Strain histograms and ratios were assessed, and areas- under-the-receiver-operating-characteristic-curve (AUROC) for each method calculated. In BI-RADS 3 tumours cut-offs for strain histogram and ratio values were calculated to see if some tumours could be upgraded for immediate biopsy. Linear regression was performed to evaluate the effect of tumour depth and size, and breast density on strain elastography.

Results: Forty-four of 99 (44.4%) tumours were malignant. AUROC of BI-RADS, strain histograms and strain ratios were 0.949, 0.830 and 0.794 respectively. There was no significant difference between AUROCs of strain histograms and strain ratios (P = 0.405), while they were both inferior to BI-RADS scoring (P<0.001, P = 0.008). Four out of 26 BI-RADS 3 tumours were malignant. When cut-offs of 189 for strain histograms and 1.44 for strain ratios were used to upgrade BI-RADS 3 tumours, AUROCS were 0.961 (Strain histograms and BI-RADS) and 0.941 (Strain ratios and BI-RADS). None of them was significantly different from BI-RADS scoring alone (P = 0.249 and P = 0.414). Tumour size and depth, and breast density influenced neither strain histograms (P = 0.196, P = 0.115 and P = 0.321) nor strain ratios (P = 0.411, P = 0.596 and P = 0.321).

Conclusion: Strain histogram analyses are reliable and easy to do in breast cancer diagnosis and perform comparably to strain ratio analyses. No significant difference in AUROCs between BI-RADS scoring and elastography combined with BI-RADS scoring was found in this study.

Fig 1. Strain elastogram of a 15 mm BI-RADS 5 tumour.

The strain elastogram is displayed left of the B-mode image and shows a stiff tumour appearing mainly blue with small green areas (horizontal arrow). The tumour appears larger in the strain elastogram than in the B-mode image, indicating malignancy. The green curve on the strain quality indicator (lower right corner of the elastogram) displays the applied compression and decompression and should ideally be between the two dotted lines (red/white). The colour scale (upper left corner of the elastogram) shows the range of colours used to designate soft (red), intermediate (yellow and green) and stiff (blue) tissue. The B-mode image shows an irregular, spiculated, hypoechoic mass with posterior shadowing (horizontal arrow). Histology showed carcinoma.

Fig 2. Strain ratio calculation of a 5 mm BI-RADS 3 tumour.

The strain elastogram shows a green and blue tumour (horizontal arrow) indicating low suspicion of malignancy. ROI A is placed within the tumour and ROI B is placed in adjacent fatty tissue. The strain ratio is calculated as: average strain _{ROI B} / average strain _{ROI A.}The higher the strain ratio, the stiffer the tumour and the higher the risk of malignancy. The strain ratio, displayed in the lower right corner of the B-mode image (vertical arrow), was 1.79 signifying a relatively soft tumour. On the B-mode image, the tumour (horizontal arrow) appears oval, circumscribed, hypoechoic and wider than tall with no posterior features. Histology showed fibroadenoma.

Fig 3

(A) Strain histogram analysis of an 11 mm BI-RADS 3 tumour. The strain elastogram shows a predominantly blue tumour. Strain histogram analysis was performed from a single ROI within the tumour (ROI A). The B-mode image shows an oval, circumscribed, hypoechoic tumour, which is wider than tall and with slight posterior enhancement. Histology showed carcinoma. (B) The strain histogram is displayed as a bar chart, with the pixel colour values on the x-axis, and the number of pixels of a certain colour pr. 1000 pixels within the ROI on the y-axis. The mean colour value of the tumour is then calculated. In this case the mean colour value was 190, signifying a relatively stiff tumour, consistent with carcinoma.

Fig 4

(A) Boxplots of the distribution of strain histogram values for benign and malignant tumours. Note that strain histogram values are unitless (y-axis, range 0–255). Upper and lower box limits are the .75 and .25 fractiles respectively. The middle box line is the average of strain histogram values. Upper and lower whiskers represent .975 and .025 fractiles respectively. (B) Boxplots of the distribution of strain ratio values for benign and malignant tumours. Note that strain ratio values are unitless (y-axis, range 0-∞), and that the y-axis is logarithmic. Upper and lower box limits are the .75 and .25 fractiles respectively. The middle box line is the average of strain ratios. Upper and lower whiskers represent .975 and .025 fractiles respectively.

Fig 5. Diagrams showing the strain histogram and strain ratio values for all BI-RADS 3 tumours.

Note that strain histogram values (range 0–255) are unitless, and that strain ratios (range 0-∞) are also unitless and the y-axis is logarithmic. The dashed line on each diagram shows the cut-off used when combining strain histogram (cut-off = 189) or strain ratio (cut-off = 1.44) values with ultrasound BI-RADS scoring.