Infiltrative nature on ultrasound images is a significant feature implying a malignant breast lesion. Characterizing the infiltrative nature with high effective and computationally inexpensive features is crucial for realizing computer-aided diagnosis. In this paper, the infiltrative nature is sighted as irregularly local variance in a 1-D signal, which is induced due to the existence of some high octave energies. These energies are extractable by a modified 1-D non-recursive discrete biorthogonal wavelet transform. The experimental results show that the proposed wavelet-based features have high individual feature efficacy and the capability of improving combined feature performance.