Medical Image Analysis Based On Feature Extraction And Machine Learning

Nowadays, with the rapid development of computer technology and medicine, computer-aided diagnosis has received growing attention and is gradually showing its importance and complexity. In the study of medical imaging, X ray of breast, CT of lung nodule, type-B ultrasonic of liver, head CT and so on have achieved some results.In this paper, combining the research results at home and abroad, research firstly introduces the comprehensive theory named Bag of Keypoints to the medical image processing. Before this, Bag of Keypoints has been used in face recognition, vehicle recognition and text recognition. These fields have yielded fruitful discriminated results. By feature extraction from liver CT images, clustering and machine learning, paper completed the classification of normal liver and cancer liver. The classification of this part is based on feature points extraction and we use SIFT descriptor to describe the feature. After K-means clustering we get code book, then using support vector machine learning, we finally get the classification decision. The second classifier is based on shape feature. This part of experiment carves out the interesting area and uses Zernike to describe its shape feature. Then through support vector machine learning, we get a classification.Experiments show that, Bag of Keypoints algorithm can achieve the classification of normal and cancer liver CT image. The classification of different period in live cancer based on shape feature is also feasible. Support vector machine learning shows good generalization ability through these two experiments which have small amount of learning samples. But this method still has distance to get to the clinical application of liver CT image analysis.