Relieving photobleaching impacts on fluorescence thermometry via neural network predictions

The thermal sensitivity of luminescence intensities enables fluorescence thermometry for remote temperature probing with high spatial and temporal resolutions. However, its accuracy suffers from factors such as nonlinear thermal response and the photochemical stability of fluorescence sensors. In this work, we realized thermometric measurements with high spatial resolution at micrometer scale using thin films with europium (Eu) complexes and microscopic measurements. We identified tris(dibenzoylmethane)phenanthroline europium(III)/polystyrene (Eu(DBM) 3 Phen/PS) thin film as an optimal choice for not only its linear dependence on fluorescence intensity for temperatures of biological interest but also its stronger resistance to the photobleaching effect. More importantly, we show that the latter effect can be effectively compensated via neural network methods. This approach has been validated for surface temperature mapping at the thermal equilibrium, where better uniformity as compared with results without correcting the photobleaching effect was achieved. The temperature elevation of resistive wires due to Joule heating can be clearly identified. This work shows that neural network models are powerful tools in improving the accuracy of fluorescence thermometry and beneficial for applications ranging from biology to nanotechnologies.