Objective:

To investigate the automatic radiomics approach in predicting the association between quantitativeultrasound features and hormone receptor status in invasive breast carcinoma.

Methods:

A total of 204 patients who accepted breast cancer surgery were retrospectively reviewed f

or pre-operative ultrasound images and post-operative pathological reports. Based on the expressions of estrogen receptor (ER)

progesterone receptor (PR) and human epidermal growth factor receptor 2 (HER-2)

all patients were divided into hormone receptor positive (ER

+

PR

+

HER-2

-

) group and hormone receptor negative (ER

-

PR

-

HER-2

-

) group. Two dimensional features of shape

margin

echo pattern

posterior acoustic feature and calcification were assessed by two experienced radiologists and correlated with hormone receptor status. The same ultrasound images were then segmented for using a phase-based active contour model. The high-throughput radiomics features were extracted based on the two dimensional sonographic features. Target features were selected using Students t-test. The support vector machine classifier with radial basis function and leave-oneout-cross-validation were used to correlate quantitative sonographic features with the status of hormone receptor.

Results:

The two groups had significant differences in the objective sonographic characteristics of shape

angular/spiculated margin

echo pattern and posterior acoustic feature (

P

0.05). In the quantitative radiomics analysis

54 features were selected with high accuracy in predicting the status of hormone receptor (accuracy 67.7%

area under the curve 73.2%). In addition

in the quantitative analysis

the two groups showed significant difference in margin

echo pattern

posterior acoustic pattern and calcification (

P

0.05).

Conclusion:

The quantitative features of ultrasound radiomics were well correlated with hormone receptor status of invasive breast carcinoma. Further study is warranted to validate its value in the precise diagnosis and biological behavior prediction of breast carcinoma.