Research On The Standardize Processing Of Peripheral Nerve Internal Fascicular Groups Segmentation From MicroCT Images

Peripheral nerve injury can result in severe paralysis and dysfunction,It is a common clinical disease.Detailed intraneural spatial information can be provided by the 3D visualization of fascicular groups in the peripheral nerv e.Based on this information,Suitable surgical methods can be selected to repair clinical peripheral nerve defects.It is possible to restore the conduction and control functions of the nerve.Therefore,3D visualization technology of fascicular groups in peripheral nerve has been a research hotspot in this field.The contour information of peripheral nerve Micro CT image is the basis of peripheral nerve 3D reconstruction and visualization.The foreground and background of these images were relatively different when the images came from samples stained by different methods,such as dyed or not dyed with calcium chloride.If previous segmentation approaches were used to extract the contours of fascicular groups,then various labor-intensive feature extracting and recognition methods had to be applied,and the results were inconsistent.Thus,this paper focuses on two types of peripheral nerve Micro CT images obtained by dyed or not dyed with calcium chloride.Study how to perform a standardized segmentation process of fascicular groups.In the standardize processing,the Micro CT images from different dyed samples are processed,which results in automatically extracted contours of fascicular groups.This paper is supported by Natural Science Foundation of Guangdong Province,China(No.2018A0303130137),Opening Project of Key Laboratory of High Performance Computation of Guangdong Province,China(No.TH1528)and National Natural Science Foundation of China(No.61975248).The main contributions of this paper are as follows:1.In the previous research,different methods are needed to obtain the contour of peripheral nerve tract in different staining methods.Thus,this study proposes a standardize processing of fascicular groups segmentation from Micro CT images of peripheral nerve to extract the contours of fascicular groups in same framework.The standardize processing can adapt the images of various sizes.2.The image datasets are constructed.The images came from samples stained by different methods,such as dyed or not dyed with calcium chloride.Several key preparations are performed,such as image annotation and grouping.The dataset of images have three subsets,namely,nondyed image dataset(subset 1),calcium chloride-dyed image dataset(subset 2),and mixed image dataset(subset 3).3.Improve the network structure of the feature extraction part of the original Mask R-CNN algorithm.The network architecture with dense connection is proposed to abstract the feature of fascicular groups,and realize the transfe r between multi-layer features,and fully reuse the extracted features.Called the improved 1 algorithm.4.Based on the improved 1 algorithm,the regulation of proposal box scores in object detection part of the Mask R-CNN algorithm was improved.To reduce the missed detection rate of the algorithm.Called the improved 2 algorithm.5.Based on the improved 2 algorithm,the transfer learning strategy was combined with the Mask R-CNN algorithm in training process.In order to improve the accuracy of fascicular groups segmentation.Called the improved 3 algorithm.6.A segmentation precision threshold is proposed to evaluate the segmentation accuracy.And explored the best values of the precision threshold in the process of obtaining contours of fascicular groups from Micro CT images of peripheral nerve.The experimental results show that:(1)The standardize processing proposed in this paper has good segmentation effect for the Micro CT images of peripheral nerve obtained by two staining methods.The pixel average precision exceeds 83%,and the intersection over union exceeds 87%;(2)The contour features of the Micro CT images of peripheral nerve obtained by not dyed staining method are complex,three improvement measures can gradually improve the pixel average precision and the intersection over union.And the improved 3 algorithm is the most effective,It has fast convergence speed,good training effect,high accuracy,and is most suitable for the task that obtaining the contour of fascicular groups from Micro CT images of peripheral nerve obtained by not dyed staining method;(3)The accuracy and the ratio of intersection and union of the four algorithms are more than 90% in the peripheral nerve micro CT images which are obtained by saturated calcium chloride staining with a small number of images in the training set but with a certain regularity of the nerve bundle contour,and they have a certain stability for the change of the fineness threshold.Among them,the improved 1 algorithm has the highest intersection and union ratio when the fineness threshold is high,and has better fineness of contour acquisition;(4)In the values of the precision threshold selection experiment,The average precision and the intersection over union curves of the improved 3 algorithm intersect when the values of the precision threshold is 0.85.At this time,we can balance the average precision and the intersection over union better.The discovery is the first time.It can be seen that for the peripheral nerve micro CT images obtai ned by different staining methods,the standardize processing in this paper can accurately,rapidly,and automatically extract the contours of fascicular groups.