Microwave Breast Cancer Detection

Early detection of breast cancer significantly increases the chance of recovery. Conventional screening methods including x-ray mammography have several drawbacks such as ionizing radiation, uncomfortable breast compression, and a relatively high miss rate. Our research t‚Äčeam is developing a microwave-based breast cancer detection system as a low-cost screening tool that can complement existing technologies.  The system uses an antenna array to propagate low-power microwave signals into the breast and records the scattered waveforms in order to detect malignancies. We develop various machine learning techniques to process the recorded signals and to decide whether they correspond to a healthy breast.

 

Our research direction on statistical signal processing aspects of microwave breast cancer detection include:

  • Machine learning techniques
  • Imaging-based detection algorithms
  • Anomaly detection algorithms
  • Experimental studies

We have also finished the first round of clinical trials, please click here to see more project details.

 

 

 

 

 

 

 

Classification paradigm:

  • Feature extraction
  • Cost-sensitive support vector machine
  • Neyman-Pearson classification

We developed three classification architecture to fuse information from all signals recorded by a multistatic antenna array

               Feature Fusion Architecture                            Classifier Fusion Architecture                       Ensemble Selection Architecture

 

 

 

System and data

  • 16-element antenna array time-domain system
  • Breast placed inside radome
  • A 2-4GHz pulse transmitted in turn
  • Tissue-mimcking reast phantoms with varying dielectric properties.
  • 290 tumor-free or tumor scans collected in 20 days.
  • 96 scans collected from 12 volunteers in 8 months.

 

 

 

Experiment results

Performance comparison with delay-multiply-and-sum (DMAS)-based and generalized likelihood ratio test (GLRT)-based algorithm:

                                          Breast phantom data                                                                               Clinical trial data

           

Faculty


Prof. Mark Coates

Prof. Milica Popovic 

 

Doctoral Candidates


Yunpeng Li

Adam Santorrelli

Lena Kranold

 

Master's students


Karim El Hallaoui

Pragyan Hazarika

 

Collaborators


Emily Porter (now Postdoc at NUI-Galway)

Hongchao Song (visting Ph.D. student at BUPT)

 

                    

1. Li, Y.E. PorterA. SantorelliM. Popovic, and M. J. Coates"Microwave Breast Cancer Detection via Cost-sensitive Ensemble Classifiers: Phantom and Patient Investigation",Biomedical Signal Processing and Control, vol. 31, 01/2017.

2. Li, Y.A. Santorelli, and M. J. Coates"Comparison of microwave breast cancer detection results with breast phantom data and clinical trial data: varying the number of antennas", European Conf. Antennas and Propag. (EuCAP), Davos, Switzerland, 04/2016.

3. Li, Y.A. SantorelliO. Laforest, and M. J. Coates"Cost-sensitive ensemble classifiers for microwave breast cancer detection", IEEE Int. Conf. Acoustics Speech and Sig. Proc. (ICASSP), Brisbane, Australia, 04/2015.

4. Li, Y.E. Porter, and M. J. Coates"Imaging-based Classification Algorithms on Clinical Trial Data with Injected Tumour Responses", European Conf. Antennas and Propag. (EuCAP), Lisbon, Portugal, 04/2015.

5. Santorelli, A.Y. LiE. PorterM. Popovic, and M. J. Coates"Investigation of classification algorithms for a prototype microwave breast cancer monitor", European Conf. Antennas and Propag. (EuCAP), The Hague, The Netherlands, Apr. 2014.