Design And Implementation Of Thyroid Nodules Detection System In Ultrasound Images

In recent years,the incidence of thyroid nodules has continued to rise,seriously endangering people’s health.Ultrasound is the preferred method for clinical detection of thyroid nodules,which is relatively safe,low-cost,noninvasive and real-time display.However,ultrasound images are different from natural images.There are problems such as low contrast,poor definition,inconspicuous nodule features,and difficulty in finding early nodules,which will affect the detection accuracy.Therefore,in view of the above problems,this dissertation studies on the theoretical basis of the existing object detection technology,and proposes a nodule detection method suitable for thyroid ultrasound imaging.Firstly,this dissertation adopts the unsharp masking(USM)to preprocess the thyroid ultrasound image,which can enhance the nodule information and at the same time highlight the object contour,thereby achieving the effect of local enhancement.In order to solve the problem of lack of large-scale ultrasound image data of thyroid nodules with labeled information,rotation and mirror flip methods are used to expand the thyroid ultrasound image data set,and the original data is expanded by two times.Secondly,this dissertation proposes a nodule detection method in thyroid ultrasound images based on deep learning.By improving the YOLOX-M algorithm to solve the problems of nodule detection in thyroid ultrasound images.In order to obtain good shallow features,this dissertation introduces involution is introduced into the backbone network to improve the spatial specificity of features and provide rich details for subsequent convolution.In order to improve the detection accuracy of early nodules,deformable convolution is introduced in the S3 stage of the backbone network to flexibly handle early nodules with different geometric shapes,extract more accurate features of early nodules,and then improve the detection effect of early nodules.In order to alleviate the impact of complex background information in thyroid ultrasound images,an attention mechanism is introduced between the backbone network and the feature pyramid network to suppress redundant information,obtain information that is more relevant to the detected object,and then improve the overall detection effect.Comparative experiment illustrates that the m AP of the proposed method reaches 0.757,and the m AP@S reaches0.670,which is improved by 3.5% and 1.8% compared with the original YOLO-M.And The ablation experiment results further prove the effectiveness of the strategy used in this dissertation in dealing with the problems in thyroid ultrasound.Finally,this dissertation develops a nodule detection system for thyroid ultrasound images based on the proposed method.After the requirements analysis,the system is designed.Through functional testing,the system can accurately detect thyroid nodules of different scales.The system operation process is convenient and rapid in operation,which can meet the actual needs of different users.