Nonmodel-based bioluminescence tomography using a machine-learning reconstruction strategy

Bioluminescence tomography (BLT) is an effective noninvasive molecular imaging modality for in vivo tumor research in small animals. However, the quality of BLT reconstruction is limited by the simplified linear model of photon propagation. Here, we proposed a multilayer perceptron-based inverse problem simulation (IPS) method to improve the quality of in vivo tumor BLT reconstruction. Instead of solving the inverse problem of the simplified linear model of photon propagation, the IPS method directly fits the nonlinear relationship between an object surface optical density and its internal biolumines-cent source. Both simulation and orthotopic glioma BLT reconstruction experiments demonstrated that IPS greatly improved the reconstruction quality compared with the con-ventional approach.