Ultrarobust and Precise Luminescence Thermometry Enabled by the Combination of Reassembled Emission Spectra With Denoising Neural Network

Nanomaterial-based luminescence thermometry enables non-invasive in vivo temperature measurement with high spatial resolution, which is crucial for driving advancement in diagnostic and therapeutic technologies. However, spectral distortions and luminescence signal attenuation resulting from complex light-tissue interactions pose substantial challenges to the practical application of this method. Here, a new strategy is presented, termed reassembled emission spectra (RaES) thermometry, for ultrarobust thermal sensing in biological environments. RaES integrates the temperature-sensitive features of sub-spectra from multiple luminescent centers, creating a thermometric parameter that is exclusively governed by temperature. To enhance accuracy further, deep learning-based denoising is preliminarily incorporated into luminescence thermometry. A U-shaped convolutional neural network model with high performance is constructed with data augmentation to recover emission spectra from significant noise with minimal bias. Empowered by the denoising model, the proposed sensing approach achieves excellent results even in challenging experiments, such as temperature measurements under static blood solution interference (ΔT = 0.23 °C) and real-time thermal monitoring during dynamic blood diffusion (ΔT = 0.37 °C), where the conventional luminescence sensing method proves completely ineffective. Being independent of specific materials and equipment, this thermometry approach offers a versatile solution adaptable to harsh environments.