A diagnostic test of two-dimensional ultrasonic feature extraction based on artificial intelligence combined with blood flow Adler classification and contrast-enhanced ultrasound for predicting -positive breast cancer.

BACKGROUND

Human epidermal growth factor receptor 2 () was an important driver gene for breast cancer which had high degree of malignancy and poor prognosis. Ultrasonography was an important imaging method for the diagnosis of breast cancer, but its diagnostic efficacy of -positive breast cancer was not satisfactory. To assess the predictive value of two-dimensional ultrasonic feature extraction based on artificial intelligence (AI) combined with blood flow Adler classification and contrast-enhanced ultrasound (CEUS) for -positive breast cancer, we compared the value of the area under the receiver operating characteristic (ROC) curve (AUC) of the combined diagnosis model and single-factor models.

METHODS

A retrospective analysis was performed on 140 patients (88 -positive and 52 -negative). These patients were divided into internal test samples and external validation samples in a ratio of 7:3 randomly. The two samples were divided into -positive group and -negative group. All the patients were examined by two-dimensional ultrasound, color Doppler ultrasound, and CEUS, and AI was used to extract two-dimensional ultrasonic image features. Features of two-dimensional ultrasound included not parallel to the skin, irregular shape, unclear boundary, posterior echo attenuated, solid or cystic-solid mixed, microcalcification or coarse calcification were treated as -positive. Levels of Doppler ultrasound included level 3 and level 4 were treated as -positive. Features of CEUS included high enhancement, fast forward, centrifugal or diffuse, uneven, lesion range increased after CEUS, with perforating branches, unclear nodule boundary after CEUS were treated as -positive. The ultrasonography characteristics in different ultrasonography methods were analyzed, the parameters with statistically significant differences between groups of internal test samples were incorporated to establish a joint diagnosis model. The sensitivity, specificity and accuracy of the combined diagnosis model and single-factor models were calculated, the ROC curve was drawn to evaluate the diagnostic efficacy of the combined diagnosis model.

RESULTS

Long diameter direction, Adler grade of blood flow, contrast agent distribution characteristics, and nodule boundary after CEUS were statistically significant different between the positive and negative groups in internal test and external validation samples (P<0.05). The sensitivity, specificity, accuracy of the combined diagnosis model were significantly higher than single-parameter diagnosis method both in internal test and external validation samples, and the kappa values of combined diagnosis model were highest. The AUC of the combined diagnosis model of internal test and external validation samples was 0.861 and 0.969, which was significantly higher (P<0.05) than that in the long diameter direction (0.717 and 0.732), blood flow Adler grade (0.674 and 0.786), CEUS distribution characteristics (0.666 and 0.750), and the nodule boundary after CEUS (0.684 and 0.786).

CONCLUSIONS

The combined diagnosis model based on two-dimensional ultrasonic feature extraction, blood flow, and CEUS can effectively predict the expression of in breast cancer.