SBIR-STTR Award

The applicants will develop statistical SURFACE fractal dimension (S-fd) features, which will discriminate benign from malignant breast masses on MRI and mammographic images. S-fd features derived from three functional representations of breast mass image data will be evaluated: (1) signal intensity of mass on single MRI slice; (2) mammographic density of mass on digitized mammogram; (3) thickness of mass, computed from 3-dim MRI data. The S-fd features are statistics from Fractal Interpolation Function Models (FIFM) of breast mass image data. In prior research, FIFM BORDER fd (B-fd) features were shown to provide more robust discrimination in data-limited applications such as breast mass analysis than other fd algorithms. FIFM SURFACE fractals represent multiresolution differences between benign and malignant masses more accurately and more extensively than FIFM BORDER fractals, and therefore may provide more reliable discriminatory information. Robust S-fd features, which discriminate benign from malignant masses, will have application in computer-aided-diagnosis systems under development. PROPOSED COMMERCIAL APPLICATION: The new features will have significant value to the diagnostician who must distinguish benign from malignant breast lesions. The algorithm is readily integrated into CAD systems and has potential utility for a variety of medical and industrial applications in texture analysis of data-limited surfaces.

The overall goal of this project is to develop, evaluate and commercialize a computer-aided-diagnosis (CAD) system that supports an enhanced decision model for interpreting breast MR images. The proposed system will enable clinicians to more readily incorporate breast MR into their decision-making and patient care. The specific aim is to develop an integrated methodology to improve sensitivity rates of radiologists with minimal clinical experience in breast MR to levels that approximate sensitivity rates of radiologists with extensive experience in breast MR. The integrated methodology has three components: 1) Enhance the interpretation model previously proposed by Nunes, Schanll, et al. to contain additional decision nodes based on lesion density and kinetics. 2) Provide readers with a CAD system that displays quantitative and visualization aids to assist in the interpretation at the decision nodes. 3) Generate training tools and case studies for teaching radiologists how to interpret and identify lesion characteristics. The CAD system is based on the statistical surface fractal features and statistical border fractal features developed in prior research. The CAD system will be developed and tested using MR image data from a consortium of institutions that use different protocols and MR systems. A reader study will evaluate changes in interpretive and diagnostic performance of radiologists who have minimal clinical experience interpreting breast MR when they are provided with CAD support.

Thesaurus Terms:
breast neoplasm /cancer diagnosis, computer assisted diagnosis, computer system design /evaluation, decision making, image processing, mathematical model, mathematics, statistics /biometry breast neoplasm, computer simulation bioengineering /biomedical engineering, bioimaging /biomedical imaging, clinical research, female, functional magnetic resonance imaging, human data, mammography, women's health