Study On Single-photon Fluorescence Lifetime Imaging Technology Based On Neural Network Retrieval Algorithm

Fluorescence Lifetime Imaging(FLIM)has shown good application potential in biomedical research and clinical diagnosis due to its non-invasive,non-toxic,non-ionizing radiation,and ability to determine the exact location of lesions.The CMOS fluorescence lifetime imaging detector realized by Single Photon Avalanche Diode(SPAD)has been widely researched at home and abroad due to its advantages of small size,fast response speed and high integration.However,existing SPAD-based fluorescence lifetime imaging methods have problems such as large imaging reduction error,slow imaging speed,and low fill factor.The thesis analyzes several common fluorescence lifetime imaging methods based on single-photon detection,introduces the working principle of SPAD devices,analyzes the fluorescence lifetime imaging circuit,and expounds the theoretical basis of neural networks.And in-depth research and analysis of the pixel circuit and the fluorescence lifetime retrieval algorithm are carried out for wide-range fluorescence lifetime imaging.The main research contents are as follows:(1)A high-precision,small-size pixel circuit based on time to amplitude conversion(TAC)is designed.The SPAD device is integrated in the pixel,and the gating method is used to avoid a large amount of ambient light noise,effectively improve the detection efficiency and solve the problem of slow imaging speed,making the fluorescence lifetime imaging circuit in the balanced between high precision and wide range.Simulation results show that the pixel circuit can achieve a full-scale range of 100 ns,and the time resolution and linearity can reach 39 ps and 99%,respectively.(2)A Verilog-A model of fluorescence lifetime decay photon is established.Generate exponentially distributed photons and grab system time to simulate randomly generated ambient photons throughout the TOF measurement window.Using Cadence software,the photon model is co-simulated with the photon time of flight(TOF)measurement circuit.A 32×32 array circuit is built based on the pixel circuit.The simulation results show that the co-simulation of the established photon arrival time model and the circuit realizes the function of fluorescence lifetime detection and verifies the feasibility of the TOF measurement method.(3)The fluorescence lifetime retrieval algorithm based on LSTM neural network is designed.The Pycharm software was used to build the neural network model and train the model,and the simulation comparison of the retrieval accuracy was carried out for the selection of different hyperparameters.In order to make the data set more accurate,noise interference was considered in the process of building the histogram.At the same time,the proposed LSTM algorithm,the traditional least square method and the centroid method are used to perform single-point and array retrieval simulation analysis,and finally the fluorescence lifetime retrieval simulation of a 32×32array is carried out using the neural network retrieval algorithm.The simulation shows that compared with CMM,LSTM algorithm can widen the retrieval range from 20 ns to 90 ns.Compared with LSM,the imaging speed of 32×32 array is increased by about 8 times.(4)Develop and design fluorescence lifetime imaging software,use Python to write related programs,use Py Qt5 to create a graphical user interface for easy interaction with the program,realize login,serial communication functions,fluorescence lifetime retrieval functions and the final imaging function,in the graphical user interface,to add buttons,text boxes,and other controls for users to enter parameters and view results.Finally,the function and performance of the program are verified through testing,and the program is optimized and improved to improve the stability and performance of the program.