Early identification of breast cancer is important for reducing the mortality rate. A common screening and detection technique for breast cancer is mammography. Though Mammography is now considered as the benchmark technique for the early screening and diagnosis of breast cancer, it utilizes harmful ionizing radiations, namely, X-Rays. Moreover the procedure is quite uncomfortable, painful and embarrassing for women, making it less attractive as a preventive screening tool. These disadvantages of mammography can be overcome by using ultrasound imaging technique. Ultrasound is normally considered safe and can be used in soft tissue such as breast. However the quality of the image is compromised due to the predominance of speckle noise. This paper focuses on developing an algorithm for an automatic segmentation and classification of breast lesions from ultrasound image in which the speckle noise was reduced using Tetrolet filter and breast lesions were automatically segmented by using statistical feature based active contour method. After segmentation, for the classification of breast lesion, totally 40 features such as 15 textural, 21 morphological and 4 fractal features were extracted from the images. Optimal features were selected to increase the classification performance by using ReleifF algorithm and 10 best features were taken into account for feature ranking. These features were used to classify the lesions from breast ultrasound images by using Support Vector Machine (SVM) with polynomial kernel for the combination of texture, morphological and fractal features from the Tetrolet filter. This method would help the radiologist to detect and classify the lesions automatically.