Breast Tumor Identification in Ultrasound Images Using the Normalized Cuts with Partial Grouping Constraints

Breast cancer is one of the leading causes of death in women. Early detection and treatment plays an important role in decreasing the mortality rate. Ultrasound is increasingly been used as an adjunct to mammography to detect breast tumors. An accurate and reliable tool to assist physicians in making diagnosis and treatment decisions is essential and necessary. We propose the use of the normalized cuts (Shi, 2000) with partial grouping constraints (Yu and Shi, 2004) to segment breast tumors in cropped ultrasonic images. The experimental results indicated that the proposed approach was robust in identifying over 500 breast tumors in a wide variety of cancer types. We found that more than 90% of segmented tumors satisfied the clinical demand. We are developing it as an assistant tool to facilitate physicians in identifying breast tumors in ultrasonic images.