Graphics Process Unit For Data Processing In Dual Band Optical Coherence Tomography

Vascular Targeted Photodynamic Therapy(V-PDT)is a new type of therapy,which can selectively block the diseased vessels and achieve targeted therapy through photodynamic response,has been widely used for clinical treatments.In order to evaluate the efficacy of V-PDT and adjust the dose of V-PDT in time,it is necessary to monitor the changes of key parameters in the process of V-PDT treatment in real time.Optical coherence Tomography(OCT)is a non-destructive,fast,non-invasive,high-resolution optical imaging method,which has shown great potential in the V-PDT treatment monitoring process.Moreover,it can simultaneously monitor the two parameters of vessel morphology and blood oxygen saturation based on the dual band OCT system.Compared with other dual-mode optical monitoring systems,the system structure is simple,and it makes up for the shortcomings of single band OCT imaging in dual parameter monitoring.With the rapid development of OCT technology and ultra-high speed linear array CMOS camera,the signal acquisition speed of OCT system is getting higher and higher.Some studies show that the acquisition speed of FD-OCT system can reach 40 MHz/s.However,the relative OCT data processing algorithm is more complex,so the data processing speed of OCT system is difficult to match its data acquisition speed,resulting in the low clinical practicability.At the same time,the spectrum data collected by the dual band OCT system is twice that of the single band OCT system.Therefore,the dual band OCT system puts forward higher requirements for the data processing speed.How to improve the data processing speed to meet the requirements of real-time imaging has become the key problem for the dual band OCT system to monitor V-PDT dose.The traditional serial data processing algorithm program based on Central Processing Unit(CPU)can not meet the internal needs of real-time imaging in improving the processing speed of OCT data.Therefore,this paper introduces the concept of parallel computing of Graphics Process Unit(GPU),and uses the Computed United Device Architecture(CUDA)develops the data processing program of dual band OCT real-time imaging to realize the parallel processing of dual band OCT data.Main work content:(1)It improves,optimizes and parallelizes the traditional OCT data processing algorithm,improves the traditional serial data processing algorithm based on CPU to the optimized parallel data processing algorithm based on GPU,optimizes the efficiency of related algorithms,reduces the calculation of intermediate parameters,and parallelizes the data processing process,finally improves the OCT data processing speed.(2)The concept of CUDA stream is used to speed up the data processing of dual band OCT system.The two CUDA task streams can process the interference spectrum data of near infrared band and visible band in dual band OCT respectively,which further improves the data processing speed of OCT image.The final experimental results show that with GTX960 M as the computing unit,one frame of B-scan(size: 2048 x 475)is processed without sacrificing the imaging quality spectral data:For the working content(1),The efficiency of its optimized interpolation algorithm and dispersion compensation algorithm is improved by nearly 7 times and nearly 7.6 times,single band OCT data processing speed based on GPU increased to 37.1 f / s,dual band OCT data processing speed based on GPU increased to 18.4 f/s.For the working content(2),using two CUDA streams to further improve GPU dual band data processing speed to 23.8 f / s.Therefore,this paper verifies the feasibility of GPU accelerating OCT data processing,and provides a basis for the future dual band OCT system to further expand angiography imaging and blood oxygen function imaging,and then realize the real-time imaging of target blood vessels and blood oxygen in the process of V-PDT,which is conducive to the accurate and personalized treatment of V-PDT.