Wideband fluorescence-based thermometry by neural network recognition: Photothermal application with 10 ns time resolution

Abstract

Neural network recognition of features of the fluorescence spectrum of a thermosensitive probe is exploited in order to achieve fluorescence-based thermometry with an accuracy of 200 mK with 100 MHz bandwidth, and with high robustness against fluctuations of the probe laser intensity used. The concept is implemented on a rhodamine B dyed mixture of copper chloride and glycerol, and the temperature dependent fluorescence is investigated in the temperature range between 234 K and 311 K. The spatial dependence of the calibrated amplitude and phase of photothermally induced temperature oscillations along the axis of the excitation laser are determined at different modulation frequencies. The spatial and frequency dependence of the extracted temperature signals is well fitted by a 1D multi-layer thermal diffusion model. In a time domain implementation of the approach, the gradual temperature rise due to the accumulation of the DC component of the heat flux supplied by repetitive laser pulses as well the immediate transient temperature evolution after each single pulse is extracted from acquired temporal sequences of fluorescence spectra induced by a CW green laser. A stroboscopic implementation of fluorescence thermometry, using a pulsed fluorescence evoking probe laser, is shown to achieve remote detection of temperature changes with a time resolution of 10 ns.