TY - ABST

T1 - Relevance of echo-structure and texture features

AU - Karemore, Gopal

AU - Mullick, Jhinuk Basu

AU - KV, Dr. Rajagopal

AU - Nielsen, Mads

AU - Chidangil, Dr. Santhosh

PY - 2010

Y1 - 2010

N2 - Aim: Echostructure is an essential parameter for the evaluation of circumscribed lesions and can be described as a texture feature on ultrasound images. Present study evaluates the possibility of distinguishing between benign and malignant breast tumors using various texture features. Materials and Methods: 58 cases of breast tumor (29 each from benign and malignant) were documented under standardized conditions using a linear array machine and 7.5 MHz transducer. In each sonographic image, ROI of tumor was marked and then subjected to the evaluation of tumor status using five parameters of second order texture statistics (second angular moment, contrast, correlation, entropy, inverse difference moment). Results: Benign tumors were fairly distinguished from malignant in the evaluation of the all texture features on the basis of wilcoxon test (P < 0.0001) but not in the second angular moment (p <0.05). Performance of individual texture feature were evaluated further on the basis of area under Receiver Operative Curve (ROC) as (0.72(p=0.0016),0.76(p=0.0001),0.72(p=0.0007),0.74(p=0.0002),0.72(p=0.0008)) respectively. Correlation between various textures features are computed using pearson’s coefficient. Best performance ROC= 0.78(p<0.0001) were obtained after combining all features using logistic regression with accuracy of 70.69 %. Conclusion: This study concludes that texture analysis appears to distinguish between benign and malignant tumors. It also reveals that when evaluating images of a breast tumor not only are the shape and size but also the echo structure of the mass is important. This technique can be considered as assistance to the radiologist to provide additional information in differentiation of benign from malignant findings.

AB - Aim: Echostructure is an essential parameter for the evaluation of circumscribed lesions and can be described as a texture feature on ultrasound images. Present study evaluates the possibility of distinguishing between benign and malignant breast tumors using various texture features. Materials and Methods: 58 cases of breast tumor (29 each from benign and malignant) were documented under standardized conditions using a linear array machine and 7.5 MHz transducer. In each sonographic image, ROI of tumor was marked and then subjected to the evaluation of tumor status using five parameters of second order texture statistics (second angular moment, contrast, correlation, entropy, inverse difference moment). Results: Benign tumors were fairly distinguished from malignant in the evaluation of the all texture features on the basis of wilcoxon test (P < 0.0001) but not in the second angular moment (p <0.05). Performance of individual texture feature were evaluated further on the basis of area under Receiver Operative Curve (ROC) as (0.72(p=0.0016),0.76(p=0.0001),0.72(p=0.0007),0.74(p=0.0002),0.72(p=0.0008)) respectively. Correlation between various textures features are computed using pearson’s coefficient. Best performance ROC= 0.78(p<0.0001) were obtained after combining all features using logistic regression with accuracy of 70.69 %. Conclusion: This study concludes that texture analysis appears to distinguish between benign and malignant tumors. It also reveals that when evaluating images of a breast tumor not only are the shape and size but also the echo structure of the mass is important. This technique can be considered as assistance to the radiologist to provide additional information in differentiation of benign from malignant findings.

M3 - Conference abstract in proceedings

BT - Euroson 2010: 22nd Congress of EFSUMB : 10th International Congress on Interventional Ultrasound, 22-25 August 2010, Copenhagen

Y2 - 20 August 2010 through 25 August 2010

ER -