Study On DT-Hessian Enhancement Algorithm For 3D Images Of Neuron

The brain consists of a large number of neurons with complex morphology and different types.Neurons are the basic units of brain structure and functions.Understanding the structure of neurons and their connections is helpful to the further study of brain functions and diseases.With the development of optical microscopic imaging,it is possible to acquire the 3D spatial morphological data of neurons.And so,the researchers can quantify neuron morphology and connections by tracing and reconstructing them.However,there are some problems in the 3D optical images of neurons,such as weak signal,strong noise,and uneven signal distribution.All these problems seriously restrict neuron tracing and reconstruction.Although many excellent achievements have been made,3D image enhancement of neurons has remained a hot topic in neuroscience.As we know,the existing methods of enhancement may have problems:poor compatibility between neuron fibers and soma;time-consuming which hinders the application of massive images.This thesis proposes an adaptive enhancement for the 3D image of neurons based on distance transform.First,the window size of the Hessian matrix is determined by distance transform:distance transform can measure the thickness of different neuron fibers effectively,and it can be used to calculate the Hessian matrix.Second,the method constructs a structural response function to enhance the neuron fibers:when the image signal is bright and the background is dark,the Hessian matrix eigenvalue?₁ is approximately equal to 0,?₂ and?₃ are less than 0 and their amplitudes are close to each other.The method takes advantage of this characteristic to enhance the neuron fibers.Third,the method enhances the soma with different strategy:the signal value of the soma is high and the radius is large.Accordingly,this method sets a higher threshold for distance transformation to localize the soma for hole filling.Through the above three steps,the enhancement of neuron with different structures is finally achieved.To evaluate the performance,this thesis applies the algorithm to enhance a variety of data and compares it with the previous algorithms.First,the results of different algorithms for synthetic data show that the performance of the proposed algorithm is better than the others.Second,testing with real neuron images,the proposed algorithm can preserve more image details.Third,the time consumption of the proposed algorithm is found as the minimum among all the algorithms for enhancing the same data.At last,the automatic tracing effect of neuron enhancement images is evaluated.Tested on two neuron datasets,the proposed algorithm can provide better input images for automatic tracing of neurons,but there is still some gap between ground truth and tracing result.To efficiently extract detailed neuron morphological information in the whole brain,this thesis proposes a basic enhancement procedure for the relatively larger regions containing neurons.First,the image regions containing neurons are manually selected from the whole-brain dataset.Second,the selected images are preprocessed and the parameters of the algorithm are unified in subsequent algorithms.Finally,the data is processed block by block with a sliding window.By executing this procedure,this thesis demonstrates the enhancement of two datasets of mouse brain.In fact,this algorithm can be extended to whole-brain data at TB-level scale.In summary,the enhancement algorithm for 3D neuron images based on distance transform can preserve the neuron signals of subtle intensity.The enhanced results have shown better visual effect,higher image quality,more abundant information.In addition,the proposed algorithm has shorter time consumption.Thus,this algorithm will benefit the morphological study of neurons.