The case for neuro-backed nano-structured artificial noses for breath biopsy

The case for neuro-backed nano-structured artificial noses for breath biopsy

Mecit Yaman*, University of the Turkish Aeronautical Association, Ankara, Turkey

Full Poster:

http://ow.ly/EzAV30mtfRF

Abstract:

Deep learning besides many other enabling uses is also beginning to emerge as a supplement computational platform to manage unspecificity afflicting practical nanotechnology. Here we demonstrate the enabling facility of neural computation with an on-chip and in-fiber odor sensing platform that use IR spectrometry for transduction. The use of a neural backend turns the simple filter instrument  into a high performance artificial-nose for breath biopsy. The sensor-array instrument can record a low fidelity, high redundancy photo-representation of the chemical fingerprints of breath analytes, and using for example autoencoders, one can carry out longitudinal label-free breath contents analysis -essentially transforming olfactory sensing into an image/video recognition/classification problem. In this poster, to demonstrate the extraordinary performance of the system, we investigate the breath analysis capacity of the filter system. We simulate human breath using FTIR spectra of 100 volatile organics taken from a commercial library; random noise is added with signal-to-noise ratios of 10 to 1 giving rise to large variations and ambiguities in the analyte mixture. A one-dimensional imaging array of 500 pixels with highly overlapping Gaussian responses is used. The neural-net consists of an three layer autoencoder network. The trained network was evaluated to reach accuracies of above 80 to 95% for the above signal-to-noise ratios. Tandem deep neural networks tolerates large variation, extreme noise and is virtually unaffected by filter or imaging array finesse, thus markedly removing stringent manufacturing and data collection requirements of the nanostructured artificial nose system rendering it to be useful for breath analysis.