Data Analysis In Breast Cancer Computer-aided Diagnosis On Images

Breast cancer is a common type of cancer that threatens women health.Early detection and treatment is crucial in promoting the recovery rate as well as saving patients.With the development of modern medical science and information technology,medical imaging techniques are playing an increasingly important role for the early detection and diagnosis of breast cancer.Computer-aided diagnosis,which help physicians draw conclusions about the by automated detection and recognition,is one of the key applications of medical imaging techniques.It helps solve the problem of the fatigue of physicians effectively.Furthermore,computer-aided diagnosis can also help avoid the bad effect caused by the gap of vocation skills among different physicians.Because of this,it is of high value in medical diagnosis and has also been a hot research spot in recent years.Based on some previous work,this thesis focuses on building a breast cancer computer-aided diagnosis system with machine learning and data mining methods.Three following aspects of the systems were researched in detail for a higher diagnosis accuracy:(1)Find the optimal classification algorithm for the breast cancer diagnosis task;(2)Study the widespread problem of data imbalance in medical data-sets and propose a better solution with ensemble learning algorithms to increase the performance of classifiers when there is a large gap between the size of positive and negative subset;(3)Combine statistical methods with physical meanings of the masses' features in various medical images,and then find appropriate ways to extract and select effective features for feature optimization and dimensionality reduction based on the subspace method.The test results on DDSM mammography data-set and the Magnetic Resonance Images(MRI)built by the author suggest that the proposal can achieve an ROC-AUC value higher than 0.95,which is better than the state-of-the-art methods.In a word,this thesis studies the breast cancer computer-aided diagnosis systems based on machine learning and data mining,and gets an ideal effect.The achievements of the thesis are promisingly to improve the performance of the existing system,and promote the early diagnosis accuracy of breast cancer.