Diffuse fluorescence tomography (DFT) is an emerging optical imaging tool for in-vivo observation of organisms and small-animal. Especially, dynamic diffuse fluorescence tomography can provide contrast-enhanced and comprehensive information for tumor diagnosis and staging with the pharmacokinetic image. However, the conventional reconstruction algorithms for DFT always suffers from low spatial resolution. Multi-modality imaging methods have been proposed to integrate DFT with other imaging modalities with in general intricate and costly experimental apparatus. We developed a dual-modality system that combines the ultrasound imaging and DFT which is simple and low-cost with no ionization damage. The results in phantom experiments demonstrate that with the a priori guidance of ultrasound imaging, the quantitativeness and spatial resolution of the fluorescence image can be considerably improved.
Dynamic fluorescence diffuse optical tomography (FDOT) can provide contrast-enhanced and comprehensive information for tumor diagnosis and staging. Based on a two-compartmental model, the adaptive extended Kalman filtering (EKF)for dynamic FDOT is proposed in our previous work to obtain better estimation than the conventional EKF. However, the set of the measurement error covariance matrix still affect the performance of the adaptive-EKF that is always manually adjusted. In order to quickly and accurately obtain the measurement covariance matrix, this paper uses generalized regression neural network (GRNN) to predict the value in different measurements as the initial covariance value of the adaptive-EKF. The simulation results have proved the validity of the measurement covariance matrix predicted by GRNN, and the images of fluorescence pharmacokinetic parameters can be achieved excellently using the adaptive-EKF and GRNN-learning.