SBIR-STTR Award

We propose developing a robust algorithm to evaluate the fractal dimension (fd) of borders of Magnetic Resonance (MR) images of breast masses which contain a small number of pixels. The fd algorithm will be evaluated in upcoming MR breast clinical trials and will be marketed to developers of computer-aided diagnosis systems for MR breast imaging. Contrast-enhanced MR is a promising tool for detecting and diagnosing masses in dense, radio-opaque and scarred breasts. Since border roughness is correlated to breast cancer and fd is a measure of roughness, the proposed algorithm will generate an objective indicator of malignancy. Current algorithms for estimating fd which use box counting or fractional Brownian motion are non-robust when applied to images with limited pixel data. The proposed algorithm generates a family of fractal interpolation function models and derives robust fd estimates from she statistics of the models. The proposed algorithm computes the following features: (a) an estimate of fd of the mass border, (b) a measure of the reliability of the estimate, (c) a measure of the extent of self-affinity of the border, and (d) a measure of the stability of the estimate over a range of threshold levels. PROPOSED COMMERCIAL APPLICATION: A computer-aided diagnosis system which reliably discriminates benign from malignant masses will have a significant market value to MR! centers and developers who have an interest in developing and promoting MR! as an adjunctive method of screening for breast cancer. Since our product enhances the performance of such a system, there is a large commercial potential and a readily identified pool of prospective strategic partners.

Thesaurus Terms:
breast neoplasm, breast neoplasm /cancer diagnosis, computer assisted diagnosis, computer system design /evaluation, diagnosis design /evaluation mathematical model, mathematics, method development, model design /development female, human data, magnetic resonance imagingNational Cancer Institute (NCI)

The objectives of this project are: (1) to produce a prototype which will demonstrate how a combination of analytically-determined MRI features, expert findings and patient data can aid the diagnostician in improving the management of patients with known focal breast masses, and (2) design a clinical evaluation of a commercial system. The prototype will generate likelihood estimates of malignancy using expert-observer readings, pre-test risk factors and computer-generated fractal dimension (fd) features as input to a non-linear discriminator. The fd feature is computed as statistics from a space of fractal interpolation function models (FIFM) of boundary segments of the mass. Boundary segments from multiple threshold levels are used. The statistical approach provides a robustness which is not found in other fd estimators. In the Phase I feasibility study, the combination of FIFM and expert-observer features generated improved discrimination over expert-observer features alone. PROPOSED COMMERCIAL APPLICATION: A system which is an aid to the diagnostician in improving the management of patients with known breast masses will have a significant market value to MRI centers and developers who have an interest in developing and promoting MRI for breast cancer diagnosis.