Wavelet-based channel energy with low cost and high efficacy is a valuable feature for the differential diagnoses between benign and malignant breast lesions. The new feature is a contour approach that generally suffers from lacking a reliable contour detection algorithm with convincing results due to extreme noise. For investigating a procedure suitable for clinical application, noise resistance capability of the new feature is evaluated in this study. The evaluation system consists of two snake-based contour detection algorithms associated with two pre-processes. These combinations can produce four test datasets of contour sonogram. Classification performance evaluation is based on a probabilistic neural network and a genetic algorithm used for distribution parameter determination.