Fully Automatic Segmentation Research Of Thyroid Nodules In Ultrasound Images

Recently,the incidence of thyroid nodule is gradually increasing.In many thyroid imaging diagnostic techniques(ultrasound,MRI,X-ray,CT),ultrasound technology is becoming more and more important because of low cost,short fetch time and so on.In traditional manual diagnosis,the doctors usually make diagnoses by shape,echo,and calcification features,which often leads to misidentification.Tissue slice has more accurate results but the workload is large,and easy to give the patients more psychological burden.With the development of image processing and pattern recognition technology,the computer aided diagnosis technology is becoming more and more important.In thyroid nodule CAD system,nodule segmentation plays an important role,which can provide accurate nodule position and shape information for doctors and firm foundation for subsequent diagnosis.However,the nodule segmentation task is more challenging due to the severe presence of speckle noise,intensity heterogeneity and low contrast in B ultrasound images.To avoid manual intervention and improve the diagnostic efficiency,we present a fully automatic thyroid nodule segmentation method.Our method includes nodule detection and nodule segmentation.The detection work locate the nodule by a target frame and the segmentation realize the pixel-level classification result.To solve the adhesion problem of the nodule and surrounding tissues met in nodule detection,we propose a nodule detection method based on edge-guided separation.First,acquire the region segmentation map by patch classification and the phase asymmetry feature map,then design edge-guided separation to solve the adhesion problem by use of the advantage of phase information in edge detection,design the nodule region selection strategy to get the accurate and complete thyroid nodule region.To solve the boundary leakage and local minimum problem met in nodule segmentation,we propose a nodule segmentation method based on neutrosophic theory and phase asymmetry.The initial evolution curve of proposed segmentation model is calculated by the nodule detection result.Structure the fusion image by neutrosophic modeling,enhance the phase asymmetry feature map of the fusion image by selective enhancing method based on two-step Otsu,design the speed stopping function to structure the thyroid nodule segmentation model based on distance regularization model.Experiment results verify the validity.This paper discusses a effective thyroid nodule fully automatic segmentation method in ultrasound image.The experiment results show the validity of the detection and segmentation,our work can purpose a reliable evidence for the subsequent thyroid nodule classification work.