OBJECTIVE: To evaluate the possibility of distinguishing between benign and malignant breast tumors using a computer-aided evaluation of echogenicity and echostructure of ultrasound findings at certain focal points. STUDY DESIGN: The ultrasound images from 89 cases of breast tumor were documented under standardized conditions using a linear array machine and 7.5 MHz transducer. In each sonographic image, the maximum area of the 'region of interest' of the tumor was marked and then subjected to consecutive statistical analysis and correlation with the histological findings. For evaluation of tumor status eight parameters of first and second order texture statistics (gray level histogram, Fourier analysis, co-occurrence matrix) were applied. RESULTS: Benign tumors were clearly distinguished from carcinomas in the evaluation of the co-occurrence matrix and the Fourier analysis on the basis of Wilcoxon and Student t-test (P < 0.05) but not in the gray level histogram. Using logistic regression a sensitivity of 73.8% and a specificity of 54.2% were obtained. A statistically significant difference between benign tumors and moderately differentiated together with poorly differentiated carcinomas could be demonstrated. CONCLUSION: This study concludes that texture analysis appears to distinguish between benign and most malignant tumors. A computer texture analyzing system is able to improve the subjective assessment of ultrasound images of the breast but can not replace it. Where the limits of subjective assessment of a given tumor are reached, computerized texture analysis will provide additional information in the differentiation of benign from malignant findings.
OBJECTIVE: To evaluate the possibility of distinguishing between benign and malignant breast tumors using a computer-aided evaluation of echogenicity and echostructure of ultrasound findings at certain focal points. STUDY DESIGN: The ultrasound images from 89 cases of breast tumor were documented under standardized conditions using a linear array machine and 7.5 MHz transducer. In each sonographic image, the maximum area of the 'region of interest' of the tumor was marked and then subjected to consecutive statistical analysis and correlation with the histological findings. For evaluation of tumor status eight parameters of first and second order texture statistics (gray level histogram, Fourier analysis, co-occurrence matrix) were applied. RESULTS:Benign tumors were clearly distinguished from carcinomas in the evaluation of the co-occurrence matrix and the Fourier analysis on the basis of Wilcoxon and Student t-test (P < 0.05) but not in the gray level histogram. Using logistic regression a sensitivity of 73.8% and a specificity of 54.2% were obtained. A statistically significant difference between benign tumors and moderately differentiated together with poorly differentiated carcinomas could be demonstrated. CONCLUSION: This study concludes that texture analysis appears to distinguish between benign and most malignant tumors. A computer texture analyzing system is able to improve the subjective assessment of ultrasound images of the breast but can not replace it. Where the limits of subjective assessment of a given tumor are reached, computerized texture analysis will provide additional information in the differentiation of benign from malignant findings.
Authors: Jimena Olveres; Germán González; Fabian Torres; José Carlos Moreno-Tagle; Erik Carbajal-Degante; Alejandro Valencia-Rodríguez; Nahum Méndez-Sánchez; Boris Escalante-Ramírez Journal: Quant Imaging Med Surg Date: 2021-08