SVM classification of breast tumors on ultrasound images using morphological features



This work aims at investigating morphological features in distinguishing malignant and benign breast tumours on ultrasound images. Support Vector Machines were applied as the classification methodology. Performances were assessed with accuracy, sensitivity and specificity. As previously seen, the most relevant individual feature is the normalized residual value, calculated from the convex polygon technique. When combined, normalized residual value, morphological-closing ratio and overlap ratio achieved an accuracy of 87% in distinguishing malignant and benign breast. The methodology developed here can be used to analyse other databases.


breast cancer,classification,support vector machines,ultrasound


Congresso Brasileiro de Engenharia Biomédica (CBEB) 2012

Cited by

No citations found