3D Segmentation And Quantitative Analysis Of Vascular Photoacoustic Images Based On Hessian Matrix

The changes of microvascular structure around tissues can reflect the occurrence and development of many vascular diseases.During the growth,invasion and metastasis of tumors,a large number of capillaries with abnormal morphology and structure and complex network distribution are generated.In this study,Optical Resolution Photoacoustic Microscopy(OR-PAM)was used to capture microvascular images of early tumors in mice.For the deep and dense network of small blood vessels around the tissue,the scattering of light or the lack of stability of the imaging system during the imaging process leads to blurred areas,noise and burr signals at the edges of the blood vessels.Thus,this paper proposes a vascular segmentation and microvascular quantification method applicable to photoacoustic 3D microvascular images in order to enhance the vascular network and obtain vascular structure parameters.It mainly focuses on the following two points:First,pre-processing techniques to remove image noise and improve contrast are utilised,which can highlight more microvascular structures.Previously based on Hessian matrix eigenvalue analysis,which can lead to breaks in segmentation at vessel junctions or intersections,in this paper,a multi-scale anisotropic tensor approach was used to obtain continuous vessel structures.In contrast,for the less-than-complete faint vessel networks with large differences in vessel diameter or low contrast,the segmented continuous vessels were processed using morphological reconstruction principles to fuse the segmented continuous vessels with the vessel grey-scale information obtained based on threshold segmentation in three dimensions to obtain a more comprehensive vascular network.At the same time,the segmentation results before and after improvement based on Hessian matrix eigenvalues are analysed for the vessel volume data and the intensity comparison method is used to verify the accuracy of the fusion segmentation algorithm.Next,the vascular skeleton was extracted using a multi-template fast marching method,in which the lack of accuracy of previous skeleton extraction methods was overcome.Furthermore,vessel diameter,vessel curvature,microvessel density and fractal dimension were chosen to characterise the structural parameters of the vessels,and the physical significance of each parameter was illustrated with the help of the structural distribution of the mimic data.At the same time,the differences in quantification of the structural parameters of the vessels are illustrated by means of two-dimensional and three-dimensional quantification methods.The results show a three-dimensional quantification of the structural parameters of the vessels,which can be closer to the structural distribution of real vessels.In conclusion,complete vessel segmentation and quantification of structural parameters are achieved for the photoacoustic microvascular network on 3D structures,which allows for more comprehensive and accurate vascular morphological information to be obtained.At the same time,3D quantification is of great importance in providing data support for the analysis of vascular morphological distribution of tumours and for clinical precision treatment.