Research On Micro-vessels Segmentation In Slight Defocused Microscopic Images

The morphological structure of the conjunctival micro-vessels of the human eye can reflect the severity of cardiovascular diseases such as hypertension,coronary heart disease and diabetes.By screening the morphological characteristics and quantity of the conjunctival micro-vessels,it can assist physicians to diagnose the patient's condition.This shows t hat the accuracy of micro-vessel segmentation is crucial.In clinical practice,the segmentation of bulbar conjunctival vessels is often performed manually.This method is affected by human factors and takes a long time,which cannot meet the requirements of large-scale fundus image processing.Therefore,using computer-aided to find efficient and accurate micro-vessel image segmentation algorithm and realize automatic segmentation technology of blood vessel image plays an important role in improving doctors' diagnosis efficiency of vascular diseases and reducing medical cost.At present,research scholars have proposed many methods of micro-vessel image segmentation.Although some results have been achieved,the accuracy of segmentation still needs to be further improved.Based on this,this paper proposes two micro-vessel microscopic image segmentation algorithms,unsupervised learning and supervised learning,and applies the algorithm to the micro-vessel microscopic images collected in this paper for experiment.Main tasks as follows:1.Aiming at the problems of poor continuity of blood vessels,unclear boundaries of blood vessels and tissues,and uneven illumination of the field of view in living micro-vessel microscopic images,this paper proposes a Fuzz y Clustering with Level Set Method.The algorithm first uses the fuzzy C-means clustering algorithm to segment the preprocessed micro-vessel images to obtain the initial contour of the region of interest in the image,and uses the improved level set algorithm to achieve accurate segmentation of micro-vessel microscopic images.This algorithm effectively solves the problem that the level set algorithm needs to manually initialize the contour,and improves the accuracy of the level set algorithm segmentation.The algorithm is applied to the weak virtual focal micro-vessel microscopic images collected in this paper for segmentation processing.Experimental results show that the algorithm results in this paper are ideal,with high efficiency,good noise resistance and more accurate image segmentation results.2.In order to further improve the accuracy and automation of the weak virtual focus micro-vessel microscopic image segmentation algorithm,this paper adds a dense connection network and attention mechanism to the original U-Net network,and proposes an Attention-Dense-UNet(AD-UNet)Algorithm of microvessel microscopic image segmentation.Adding densely connected networks can make full use of the feature information of each layer and effectively solve problems such as gradient diffusion.The introduction of attention mechanism can better locate the region of interest,thereby improving the efficiency of the algorithm.The algorithm of this paper is tested on the commonly used fundus image data sets DRIVE and STARE,and their accuracy reaches 0.9663 and 0.9684 respectively.The algorithm solves the problem of small blood vessel segmentation and proves the segmentation advantage of this algorithm.The AD-UNet algorithm is applied to segment the weak virtual microcapsule microscopic images collected in this paper.The segmentation results show that the algorithm can obtain ideal segmentation results.