Applications of machine learning in biophotonics and laser metrology

Abstract

Recently, optical technologies have found several applications in fields including biophotonics, precision metrology and wavelength scale sensors. However, to gather statistically relevant information and analysis these methods require large amount of measurements. Current linear multivariate methods such as principal component analysis or linear discriminant analysis are not sufficient to analyze these big datasets with non-linear variability. Recently, the application of deep learning based artificial neural networks have found an upsurge in various areas of science ranging from quantum physics to evolutionary biology, providing an enhancement in the efficiency of various techniques. This thesis focuses on the applications of machine learning with the goal to enhance different aspects of biophotonics.

Firstly, this thesis explores the application machine learning to enhance the label-free characterization of cells of the immune system using Raman spectroscopy and digital holographic microscopy. The combination of deep learning with digital holographic microscopy provides a route towards a high throughput hemogram device which would be useful for the classification of clinically important immune cells with morphological similarities but different functions.

Following this, the applications of deep learning are explored in the regime of precision optical metrology for the development of a laser speckle wavemeter with a high dynamic range with an additional application for the development of a binary speckle based spectrometer.

Finally, the application of machine learning based methods are also explored to improve the sensitivity of the chirped guided mode biosensor. A comparison between the linear method of principal component analysis and direct Fano fitting is drawn which is followed by the application of multi layered perceptron for further improvement.