The goal of this study is to design quantitative methods that clinicians can use to supplement their visual interpretation of sonograms for differentiating benign and malignant solid breast masses. The hypothesis is that combining quantitative methods with clinicians'assessment of images will improve the accuracy of diagnosis and reduce the number of false positive or unnecessary biopsies. Our preliminary study shows that certain sonographic features derived from lesion margin, shape, and echo characteristics can help differentiate benign and malignant solid masses. In this application, we propose to build on our initial success and develop a diagnostic system on an ultrasound scanner that provides the end user with online estimates of probability of malignancy from quantitative analysis of the breast ultrasound images. The program has four specific aims.
In Specific Aim 1, ultrasound images of breast masses from 400 patients will be acquired under controlled and well- defined experimental conditions.
In Specific Aim 2, new approaches will be developed to detect mass margins and to describe these features quantitatively. The qualitative features of the masses that clinicians use in routine diagnosis will also be identified. The quantitative and the qualitative feature sets will be used individually with novel classification methods based on logistic regression, neural networks and radial basis function classifiers to formulate a decision tree for cancer diagnosis. The diagnostic performance of each classification scheme and feature set will be evaluated by ROC analysis.
In Specific Aim 3, the qualitative and the quantitative feature sets will be combined, integrating the intuitive medical experience of the clinicians with the precision of quantitative measurements. In the final phase of the program, Specific Aim 4, the best performing feature set and classification scheme will be implemented on an ultrasound scanner for online diagnosis of malignant and benign breast masses. This program integrates qualitative clinical and quantitative computer approaches for breast cancer diagnosis. We expect to develop a new diagnostic system that determines probability of malignancy, which clinicians could use as an online second opinion when making diagnostic decisions during the performance of a breast ultrasound examination.

Public Health Relevance

Breast cancer is the second leading cause of cancer death in the US. Currently ultrasound imaging is used for diagnosing breast cancer by visual inspection of the images. This application introduces a new paradigm that will use quantitative methods to differentiate malignant and benign breast masses. If successful, the proposed research could reduce the number of false positive or unnecessary biopsies.

Agency
National Institute of Health (NIH)
Institute
National Cancer Institute (NCI)
Type
Research Project (R01)
Project #
5R01CA130946-05
Application #
8387042
Study Section
Medical Imaging Study Section (MEDI)
Program Officer
Baker, Houston
Project Start
2008-12-01
Project End
2014-11-30
Budget Start
2012-12-01
Budget End
2014-11-30
Support Year
5
Fiscal Year
2013
Total Cost
$297,988
Indirect Cost
$108,789
Name
University of Pennsylvania
Department
Radiation-Diagnostic/Oncology
Type
Schools of Medicine
DUNS #
042250712
City
Philadelphia
State
PA
Country
United States
Zip Code
19104
Venkatesh, Santosh S; Levenback, Benjamin J; Sultan, Laith R et al. (2015) Going beyond a First Reader: A Machine Learning Methodology for Optimizing Cost and Performance in Breast Ultrasound Diagnosis. Ultrasound Med Biol 41:3148-62
Bouzghar, Ghizlane; Levenback, Benjamin J; Sultan, Laith R et al. (2014) Bayesian probability of malignancy with BI-RADS sonographic features. J Ultrasound Med 33:641-8
Pouch, Alison M; Cary, Theodore W; Schultz, Susan M et al. (2010) In vivo noninvasive temperature measurement by B-mode ultrasound imaging. J Ultrasound Med 29:1595-606