Report

Mike Nonte Apply voltage or current with known frequency and amplitude Record current or voltage response Use phase shift and change in magnitude to determine complex impedance Sweep through a range of frequencies to produce a nyquist plot [1] Data Set ◦ EIS recordings from 106 freshly excised breast tissue samples ◦ Each sample belongs to one of six tissue types: 1. 2. 3. 4. 5. 6. Carcinoma Fibro-adenoma Mastopathy Glandular Connective Adipose Problem: use pattern classification techniques to reliably determine tissue type from EIS recordings [2] Replace ELMs with MLPs and compare computation speed and accuracy Publically available data has nine features already extracted: I0: Impedance at zero frequency PA500: Phase angle at 500kHz HFS: High-frequency slope of phase angle DA: Impedance distance between spectral ends AREA: Area under the nyquist plot A/DA: AREA normalized by DA MAX OP: Maximum of the spectrum DR: Distance between I0 and real component of the maximum frequency point ◦ P: Length of the spectral curve ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ Previous work [2] uses mutual information to rank attribute strength then tests different feature vector dimensions to determine which yields best results Only 9 feature attributes, so an exhaustive subset selection approach is slow but possible ◦ Randomly split data into equally sized testing and training sets ◦ Train a single ELM and measure classification rate with each possible set of attributes ◦ Determine optimal feature vector 0.65 0.6 0.55 Classification Rate 0.5 0.45 0.4 0.35 0.3 IO P DA DR AREA PA500 P P IO PA500 0.25 0.2 0 100 200 300 400 500 600 # Neurons in Hidden Layer 700 800 900 1000 Short-term ◦ Apply ELM outputs to multi-class SVM ◦ Replace ELMs with MLPs and compare speed and accuracy of classification Long-term ◦ Obtain larger data set to ensure generalization of results ◦ Examine new attributes that may be more useful in determining a physiological basis for observed impedance properties [1] Williams, J. C., Hippensteel, J. A., Dilgen, J., Shain, W., & Kipke, D. R. (2007). Complex impedance spectroscopy for monitoring tissue responses to inserted neural implants. Journal of neural engineering, 4(4), 410. [2] Daliri, M. R. (2013). Combining extreme learning machines using support vector machines for breast tissue classification. Computer methods in biomechanics and biomedical engineering, (ahead-of-print), 1-7.