Uniform illumination constraint enhancement and utility weighted voting fusion for ultrasonic breast lesion segmentation

The risk of breast cancer in women increases notably with age; one of every eight women is prone to get breast cancer in her lifetime. Ultrasonography is a noninvasive and painless medical imaging technique which achieves high lesion detection accuracy. However, ultrasound images are characterized by their specular nature, attenuation, speckle, shadows, and low contrast. This work proposes a six step algorithm to identify and segment lesions in ultrasound images. The proposed methodology uses radial intensity analysis followed by a uniform illumination constraint function to highlight the region of interest. The results of four segmentation techniques are fused to yield a consensus fused map. The final lesion boundary is the one that maximizes the utility function. The results show that the proposed algorithm outperforms the four tested segmentation algorithms, resulting in an average overlap of 71.3% and deviation of 2.2%with an average modification over each of the tested algorithms of 28%.