Research On Thyroid Ultrasound Images Classification Based On Improved Integration Technique

With the development of ultrasonic imaging and medical diagnostic technique, ultrasonichas become one of the main means of thyroid cancer. At present, the determination of thyroidcancer is mainly through the doctor’s qualitative discrimination of ultrasound image, butbecause of the varibility of the biological charactar of thyroid cancer and the difference ofeach big hospital diagnosis put emphasis on, the diagnostic results are susceptible to thefactors such as he doctor’s experience, level, state, and the accuracy of the diagnosis isdifficult to guarantee. Therefore, we need to establish an objective method, and give necessaryaid to physicians in diagnosis of thyroid disease.Ensemble learning is a new type of machine learning techniques, it is in classifying newinstance, integrated a number of individual classifiers, and decide the final classificationbased on the classification results are a combination of multiple classifiers. Normaly,integrated learning can get better performance than a single classifier. The paper will beintroduced to the dynamic integration technology to medical image classification problem.This paper focus on how to make use of dynamic integration on the classification ofrecognition to solve the problem of the low recognition rate of thyroid b-ultrasound in theimage classification and recognition.Aiming at the problems above, this paper makes a thorough research on quantitativefeature extraction and integration algorithm, and mainly attains the following researchachievement:1. For traditional dynamic integration algorithm cannot get stable auxiliary,we improvedthe dynamic integration based on clustering algorithm is and improved objective function ofk-means clustering algorithm, and calculation formula of the distance of ability area, wepropose a new clustering standard, increasing the difference between the clustering samplesand improving integration algorithm accuracy; At the same time, in order to increase thestability of the classification model, we apply weighted method to the dynamic integrationalgorithm and use multiple classifiers to parallel integration. 2. By analyzing thyroid B ultrasonic image, synthesizing clinical identification ofvarious features of thyroid nodules, we carry on the quantification separately and put forwardthe unique micro-calcification metric method for thyroid nodule, and finally extract circulardegree, attenuation coefficient, degree of micro-calcification nine characteristic parameters,which can best describe the behavior of nodules, as data set for thyroid disease, and providingdoctors with a relatively objective quantitative parameters.3. To measure the performance of the classifier algorithm, this paper will be based on theimproved dynamic integration method and the commonly used linear discriminant in similarstudies, the classification effect of BP neural network and SVM algorithm are compared, andthe advantage of the algorithm is proved in this paper.