Research On Free-hand 3D Ultrasound Reconstruction Algorithm Based On Linear Transducer

As a technology that can visualize the internal structure of the lesion,medical imaging plays a very important role for clinicians.In the existing medical imaging technology,ultrasound imaging is faster,safer and cheaper than magnetic resonance imaging and computed tomography imaging.Therefore,it has become a routine method of clinical diagnosis and plays an irreplaceable role in lesion diagnosis.Compared with other ultrasonic imaging techniques,freehand 3D ultrasound reconstruction has a large scanning range and is easy for doctors to operate.The existing freehand 3D ultrasound reconstruction methods have some problems and challenges.Firstly,there are few researchs on the simulation verification model needed in the design of handheld ultrasonic transducers.Secondly,when scanning a target with a fixed surface shape,density of the medium,movement of the probe,and change of the lesion position will bring reconstruction errors.Thirdly,when scanning a target with an unfixed surface shape,it is impossible to capture small changes caused by squeezing.This is mainly due to the low frame rate of the wireless ultrasonic probe.So,the 3D reconstruction effect and real-time performance are not as good as the 2D ultrasonic probe.In this thesis,based on the freehand 3D ultrasonic reconstruction algorithm of linear array transducer,the following research results are obtained:(1)In order to obtain better freehand 3D reconstruction results,the current method needs to design and fabricate a variety of transducers for testing according to different 3D reconstruction targets.This method has a long development cycle and high cost.In this thesis,the directivity of transducer arrays such as linear array,convex array and planar array is simulated and analyzed.According to the directivity of different arrays,this thesis proposes a directivity evaluation method suitable for freehand ultrasonic probe transducer.In this method,the expressions of convex array and planar array are derived by linear array.At the same time,the transducer is modeled.The directivity of different array forms can be obtained by changing the size,number,spacing and arrangement of the transducer model.According to the simulation results,a linear array transducer is designed and fabricated for the freehand 3D reconstruction algorithm proposed in this thesis.(2)A 3D ultrasonic reconstruction algorithm PSM(Probe sector matching)based on sector matching is proposed.The existing methods mainly rely on the pose information of the external device for 3D reconstruction.when scanning targets with fixed surface shape,the change of medium density and lesion positon will affect the accuracy of 3D reconstruction.PSM uses the method of establishing a sector model to solve the problem of uneven image data caused by the non-uniform motion of the probe.At the same time,PSM uses pressure information and pose information to construct non-uniform parameter curves to correct distorted 3D reconstructed images.From the experimental results of the numerical model and the real target,PSM effectively reduces the error caused by the nonuniform motion,non-uniform medium and the movement of the lesion position.(3)A sequence prediction 3D ultrasonic reconstruction algorithm SPRAO based on acousto-optic localization is proposed.If the target surface or lesion postion changes during scanning,existing 3D reconstruction algorithms will not be able to capture small changes due to the low frame rate of the wireless probe.SPRAO synthesizes high frame rate ultrasonic image sequences by interpolation and speckle decorrelation.At the same time,SPRAO combines 3D convolutional neural networks to improve computational efficiency.In the training process of the model,the long short-term memory network is used for sequence prediction,which shortens the training time.In the experimental part,the 3D reconstruction results of SPRAO are compared with other methods by using the same ultrasound image data set.The experimental results show that SPRAO not only improves the image data density,but also eliminates the cumulative error of the external positioning device.The target surface change and the small movement of the lesion position caused by squeezing are also corrected.The research results of this thesis can be used to solve the errors caused by external positioning information or target surface changes in the process of freehand 3D ultrasound imaging.At the same time,the 3D image reconstruction algorithm combining ultrasonic images,sensor data and deep learning models can also provide a reference and test platform for real-time 3D reconstruction.