INTEGRATION OF QUANTITATIVE PHASE SPATIAL AND TEMPORAL FLUCTUATIONS IN COHERENT MICROSCOPY

Abstract

The work is devoted to the consideration of a new approach to the classification of living cells, which combines spatial and temporal fluctuations, quantitative indicators of the optical thickness of the cell, obtained using phase dynamic imaging. Five different indicators were used to evaluate the various methods: accuracy, sensitivity, specificity, reliability and statistical index (AUC). Various architectures were considered: single-path ResNet, two-way ResNet and three-way ResNet. The methods of early and late fusion were also compared using various indicators: morphology, a map of spatio-temporal fluctuations and/or a two-channel input. Combining morphology and spatial-temporal fluctuations map (triple path model) improves results from 85 % (morphology only) to 89 % (combining morphology and spatial-temporal fluctuations).