Research On FMT Accuracy Improvement Method Based On Multi-wavelength Concurrent Reconstructio

As one of the most promising new optical molecular imaging technologies,Fluorescence Molecular Tomography(FMT)uses specific molecules and fluorescent probes at the molecular level to help achieve non-invasive detection of early tumors.However,FMT image reconstruction is a highly non-linear problem.At the same time,biological tissues have a high degree of heterogeneity and complex boundary conditions,leading to serious ill-posedness inverse problems.Based on this,this study proposes a multi-wavelength-based concurrent-wavelength reconstruction algorithm(CRA),which combines multi-wavelength and FMT technology,and uses multi-wavelength fluorescence emission spectra to enrich tissue optical information in tomography,and realizes the accurate reconstruction of the target anomaly with the help of deep learning finally.The main research contents of the paper include the following three parts:(1)Feature processing based on multi-wavelength tissue surface fluorescence information.Based on the photon forward transmission model and the finite element method,the accumulation of the fluorescence intensity information on the surface of the tissue is completed under the excitation light of different wavelengths.The feature data processing method based on CRA is to first perform the principal component analysis(PCA)on the fluorescence signals obtained from the two groups of different wavelengths,and then combine with the fluorescence signal obtained from the third set of wavelength as the input data of the neural network.(2)Research on the imaging adaptability of single target anomaly and double target anomaly based on multi-wavelength concurrent wavelength reconstruction algorithm.This subject has carried out simulation experiments on the image reconstruction effect of single target anomaly,double target anomaly and the noise resistance of the model.The experimental results show that the CRA proposed in this research has high accuracy and adaptability for different number,size,position and parameters of the anomalies reconstruction.(3)Image reconstruction algorithm of anomaly based on deep neural network.In this study,Stacked Auto-Encoder(SAE)deep neural network was used to reconstruct the target anomaly image and the noise resistance of the network model was also studied.Experiments show that under different signal-to-noise ratios(SNR),the correlation coefficient between the fluorescence yield value and the true fluorescence yield value of a single target reconstructed by CRA is above 0.96;in the reconstruction of double target,compared with The reconstruction effect of a single wavelength is more advantageous.Moreover,the positioning error of the CRA reconstruction of the dualtarget is less than 0.5mm,and the area error of the reconstructed target is also lower than the reconstruction result of a single wavelength.The research results in this paper conclude that,compared with the singlewavelength reconstruction method,the proposed CRA method based on multiple wavelengths can improve the quality and accuracy of fluorescence image reconstruction.The single-target anomaly and the dual-target anomaly are reconstructed separately to verify the performance of CRA,which achieves a further improvement in FMT imaging accuracy.