基于混合光传输模型和复合正则化的生物发光断层成像重建方法研究

Bioluminescence tomography (BLT) has emerged as a major imaging modality of optical molecular imaging, it has a broad prospect for biomedical application. However, further progress will be needed in refining BLT reconstruction algorithms to fully realize its potential. In the project, a great effort is devoted to developing BLT reconstruction algorithms based on hybrid light propagation model and composite regularization method. First, a hybrid light propagation model which couples diffusion approximation (DA) and simplified spherical harmonics approximation (SPn) is developed to improve the computation efficiency and the accuracy of BLT reconstruction algorithms, and the computational speed of BLT reconstruction algorithms is further accelerated through parallel execution using a graphics processing unit; Second, a composite regularization algorithm which combines sparse regularization and total variation regularization and its solver are developed to enhance the quality of reconstructed images; meanwhile, an automatic choice of regularization parameters based on model function is developed to improve robustness of BLT reconstruction algorithms; last, in vivo experiments are performed to evaluate the effectiveness of the proposed reconstruction algorithm. Compared with the single light propagation model (DA, SPn) and the single regularization method (sparse, total variation) applied in the onging project supported by Natural Science Foundation of China, newly developed reconstruction algorithm which has faster computational efficency, higher reconstruction quality and better robustness will be ensured by the hybrid light propagation model, the composite regularization, the automatic choice of regularization parameters and the technique of GPU accelaration.

生物发光断层成像(BLT)是光学分子影像的一种重要成像模态，在生物医学领域有着广泛的应用前景，但其应用潜力依赖于重建算法研究的新进展。本课题重点研究基于混合光传输模型和复合正则化的BLT重建方法。首先，研究耦合扩散近似(DA)和简化球谐近似(SPn)的混合光传输模型，并基于图形处理单元(GPU)进行并行硬件加速，提高重建算法的精度和速度；其次，研究联合稀疏正则化和总变分正则化的复合正则化方法及其求解算法，并基于模型函数进行复合正则化参数的自适应选择，以提高重建图像质量和重建算法的鲁棒性；最后，开展在体实验验证上述算法的有效性。相比于在研青年科学基金项目中应用的单一光传输模型(DA、SPn)和单一正则化方法(稀疏、总变分)，本项目的混合光传输模型、GPU加速技术、复合正则化及正则化参数的自适应选择将确保重建算法具有更快的计算速度、更高的重建质量和更好的鲁棒性，从而促进BLT技术的进一步发展。

本项目致力于生物发光断层成像重建方法研究，力求提高重建算法的计算精度、计算速度和鲁棒性。为提高重建算法的计算精度和计算速度，本项目开展了耦合DA和SPn的混合光传输模型前向问题研究，耦合的混合光传输模型可以在保证较小的计算代价下获得比较准确的计算精度，从而在计算速度和计算效率之间取得了平衡；为提高重建算法的计算速度，本项目对生物发光断层成像的前向问题开展了并行加速求解，提高了求解速度；为提高重建算法的重建精度，开展了基于复合正则化重建方法研究，通过将L1正则项和总变分正则项相结合以利用两者之间的优点，提高重建图像的质量；为提高重建算法的鲁棒性，先后提出了一种直观的正则化参数自适应选择方法、基于归一化累积周期图的正则化参数自适应选择方法以及基于模型函数的复合正则化参数选择方法；此外，本项目还基于贝耶斯压缩感知理论进行了生物发光断层成像重建方法研究，力求从重建算法自身角度提高重建算法的重建精度和计算效率。本项目研究内容对生物发光断层成像的重建算法包括前向问题和逆向问题进行了深入研究，发表学术论文15篇，申请发明专利8项，其中授权美国和中国发明专利各1项。