Fig. 3From: Photon-counting computed tomography thermometry via material decomposition and machine learningSummary of experiment results. a The fully connected neural network architecture used to non-linearly model the relationship between attenuation and temperature. The input to the network are the spectral attenuations of a material at a baseline temperature concatenated with the attenuation residuals due to heating; b Visualization for network performance for predicting temperature on 300 mmol/L CaCl2 and a milk-based protein shake. The data points are labeled in the (xx, yy) format where xx is the predicted temperature and yy is the ground truth temperature synchronously measured with a digital thermometer. The 95%CI of temperature prediction is shaded. Data from the testing samples were not included in the training dataBack to article page