Segmentation And Measurement Of The Region Of Interest In Renal Pathological Images

During the diagnosis of chronic renal disease,the pathologist needs to locate the region of interest（glomerulus,renal tubule,renal vessel）,and then determine whether there is a lesion in the region of interest.However,the location of the region of interest in the renal pathology is numerous and not fixed.Pathologists need to frequently zoom in and zoom out the multiples of the optical microscope to find the location of the region of interest,which is time-consuming and easy to miss.In addition,the size of the region of interest is still observed using pathologists’experience,quantitative analysis of certain diseases is lacked.On the other hand,renal pathological images have the characteristics of extremely high resolution and rich fine-grained details.The current algorithms have the problems of slow running speed and insufficient accuracy,which cannot be using in clinical diagnosed.Therefore,there is an urgent need for an efficient artificial intelligence method to assist pathologists in quickly locating the area of interest and automatically measuring its size.Motivation by these above problems,this paper constructs an efficient and uncompressed segmentation model of the renal pathology region of interest in high-resolution whole slide images.By simulating the pathologist’s method of locating the region of interest,the model is divided into two stages.In the first stage,a Region of interest Object Detector（ROD）is constructed to detect the region of interest in renal pathology at the box level.ROD uses the strategy of dividing the original whole slide image into multiple small images for rapid model training,and proposes a dynamic scale evaluation method to directly test the original whole slide image.The second stage segment the region of interest at the pixel level,and restore to the original image to complete the segmentation of the region of interest in kidney pathology via using the detecting results of the first stage to construct a saliency segmentation data set,In addition,the algorithm flow designed in this paper performs automatic dimensional measurement of the size of the segmented region of interest.In this study,1204 patients’renal pathological whole slide images were collected and annotated to train and test the model.In the detecting stage,the ₁F value of ROD’s detecting of the region of interest reached 0.981,and the detecting speed reached 8.58 seconds per whole slide images.The average Dice coefficient reached 0.970 in the segmentation stage.Experiments prove that the algorithm proposed in this paper has better performance and faster running speed to meet clinical needs and greatly reducing their workload by assisting doctors in quickly locating the region of interest in renal pathological images.The quantitative analysis of the size of the region of interest enables the quantification of certain renal diseases,laying a foundation for the subsequent analysis of the correlation between the size and the disease,and thus has a promoting effect on the development of quantitative biology and renal pathology.