Purpose
The Micrima MARIA® system is a Radio frequency (3-8GHz) imaging system [1] (Figure 1) that exploit dielectric contrast within breast tissue [2,3] for cancer detection.
Using a hemispherical conformal array of 60 antennas designed to fit round the shape of the breast and with antennas that look at the breast from all sides in a close packed pattern, 1770 independent channels are measured over 101 frequency points. These data are used to form a 3D focused image of scattered signals from within the breast. Raised...
Methods and materials
Phantom target materials: lesion mimicking serum and propanediol, were prepared in 10mm and 20mm blown thin walled glass bulbs on the end of slim capillaries. They were each immersed in the MARIA’s imaging volume, and radiofrequency measurement samples were taken at various position within the volume. Forty-eight (48) samples were collected for each target material.
Non-linear support vector machine (SVM) and linear discriminant analysis (LDA) classification were performed on channel data (178770 features), or the in-image focused frequency response (101 features) that were also subjected...
Results
According to Table 1, 97% accuracy advocates ‘SVM on channel data’ as the preferred solution. However learning curve data (Figure 2a) demonstrates a lack of robustness on unseen data. The error rate from training was unrealistic throughout and validation error did not reduce to training’s level, this exhibits signs of overfitting compared to other methods.
Table 1: Summary of results.
Methods
Features
Accuracy (Monte-Carlo simulation
Robustness towards unseen data (learning curves)
SVM on channel data
178770
97%
Poor
SVM on in-image focused frequency response
101...
Conclusion
Accuracy and robustness of RF-based lesion classification are demonstrated using a adipose/lesion phantom. The results show that the focussed frequency response is optimal, because it is clear that ‘SVM on channel data’ is substantially ill-posed, i.e., the problem dimension is larger than the number of data samples. Whilst any ill-posed classification should not be rejected, care should be taken to assess performance on would be unseen data.
Applying this classification technique to the entire image volume produces an alternative way to view the image data...
Personal information and conflict of interest
T. Doshi; Bristol/UK - Employee at Micrima Limited, Bristol, UK D. Gibbins; Bristol/UK - Employee at Micrima Limited, Bristol, UK L. Tsui; Bristol/UK - Employee at Micrima Limited, Bristol, UK A. W. Preece; Bristol, AVON/UK - Founder at Micrima Limited, Bristol, UK
References
Preece AW, Craddock I, Shere M, Jones L, Winton HL. MARIA M4: clinical evaluation of a prototype ultrawideband radar scanner for breast cancer detection. J. Med. Imag. 3(3), 033502 (2016), doi: 10.1117/1.JMI.3.3.033502.
Lazebnik M, McCartney L, Popovic D, Watkins CB, Lindstrom MJ, Harter J, Sewall S, Magliocco A, Booske JH, Okoniewski M, Hagness SC. A large-scale study of the ultrawideband radio wave dielectric properties of normal breast tissue obtained from reduction surgeries. Phys. Med. Biol., vol. 52(10):2637-2656, 2007.
Lazebnik M, Popovic D, McCartney L, Watkins...