Automated Staging Of The Mouse Testis Seminiferous Epithelial Cycle Based On Tissue Image Analysis

With the increasing incidence rate of infertility,infertility has become the third most difficult disease in the world,which seriously affects human reproductive health.Because of the similarity in the structure of mammalian testis,the mouse is usually used as an animal model for experiments before the pathological study of human testis.In the process of mouse spermatogenesis,the entire seminiferous epithelium cycle can be divided into 12 stages.Histologists mainly observe the morphology,spatial location and texture of testicular cells in seminiferous tubules when performing staging work.Accurate staging of the mouse spermatogenic epithelial cycle is useful for analyzing the composition of testis cells,understanding the pathogenesis of infertility,and assessing the quality of the spermatogenic process.Most histologists complete staging under the light microscope,and the judgment process is time-consuming and labor-intensive and has a certain degree of subjectivity.Therefore,it is necessary and meaningful to use computational pathology technology to develop an automatic staging model to assist doctors in staging judgment.The first work in this paper applies a deep learning approach to segment testicular cells in seminiferous tubules from H&E-stained cross-sectional panoramic tissue images of mouse testis and perform compositional analysis of testicular cells.An automatic segmentation method based on multi-head self-attention mechanism and distance map is proposed.This method introduced the multi-head self-attention mechanism,which makes the model can obtain cell position information and enhances the accuracy in cell segmentation.The distance map branch was added at the same time,whose outputs can be used as the watershed markers,solving the problem of possible overlap between cells.Finally,the model can complete the automatic segmentation of multiple testicular cells and cell foreground background simultaneously.The PA,CPA,Io U and FWIo U of the testicular multi-type cell segmentation branch of the model reached 0.962,0.909,0.833 and 0.928 on the test set.The Dice coefficient,Precision,AJI and HD distance of the cell foreground and background segmentation branch on the test set were respectively The performance is better than other models.After accurate segmentation of testicular cells,the segmentation results at stages Ⅵ to Ⅷ were not statistically significantly different from the histologist’s annotation through t-test.After completing the automatic segmentation of testicular cells,the composition of testicular cells was analyzed,and it was found that the number of round spermatozoa in the late stage of the seminiferous epithelial cycle was greatly reduced,and spermatogonia and spermatocytes were basically in the early and middle stages.The number of stages will increase,the number of late stages will decrease,and Sertoli cells do not participate in the process of sperm development,so the number is always maintained at a relatively constant level,and these changes are basically consistent with the actual biological laws.The second work of this paper is to fuse multiple features to achieve automatic staging of mouse testicular seminiferous epithelial cycle stages Ⅵ to Ⅷ.A deep model is proposed to stage the spermatogenic epithelial cycle of mice,and extract the features of the network after maximum pooling as deep features,and test them on the test set.Better.Cell morphological features Based on the segmentation results in Chapter 3,the spatial distribution features,morphological features and color texture features of cells are extracted,with a total of 417 dimensions.For the proposed features,the first 12-dimensional features were screened out by m RMR,Wilcoxon rank sum test,Relief F algorithm,Fisher algorithm,and classifier models such as SVM,LDA,QDA,KNN were constructed for staging.Among the staging models,the LDA model has the best performance.Finally,the depth features were reduced in dimension and fused with cell morphological features,and the constructed staging model ACC reached0.735.According to the 1 VS Rest strategy,in the results of 10-fold cross-validation,the AUC values of the best models are 0.81,0.80,and 0.90,respectively.Compared with histologist’s staging work,computer automatic staging has higher efficiency and accuracy,and can better predict spermatogenesis.The research in this paper is based on computational pathology technology.The two tasks can not only accurately segment testicular cells and perform large-scale statistical analysis,but also perform staging of mouse spermatogenic epithelial cycles from Ⅵ to Ⅷ,assisting histologists to make judgments.,can reduce the work pressure of histologists and promote the development of reproductive medicine.