Medical Image Computer-Aided Diagnosis Based On Kernel Matching Pursuit

With the development of information science technology, image auto-processing technology has been the indispensable important tool for science research. One representative application is medical image engineer. Medical image mining has been one of the fastest developments of medical technology, and it provides scientific foundation of clinical diagnostics and biomedical research. Based on this, the diagnosis accurate can greatly improve by machine learning methods using a prior knowledge to guide computer-aided diagnosis of the unknown image data. This paper uses machine learning into medical image diagnosis, major tasks are as follows:According to the problems of significant individual difference, few data labels and unbalanced data distribution of medical images, the medical image computer-aided diagnosis (CAD) technique, combined with several segmentation processing methods on the region of interest (ROI) based on Kernel Matching Pursuit (KMP), is proposed in this paper. The results on diagnosing medical images are improved. Furthermore, aiming to weak samples of the unbalance data, Adaptive Kernel Matching Pursuit (AKMP) based on the feedback penalized function mechanism is proposed. Experimental results show that the proposed pretreatment method increases diagnosis accuracy rate of medical images effectively. Because of the low diagnosis accuracy of whole medical image, we present a novel transfer learning model based on kernel matching pursuit called TLKMP. TLKMP uses the Greedy Approximation Residue to transfer instance into target domains which have little labeled set of different distributions from the source domains. So, valuable instances in source domains are reused to construct high quality classification for the unlabeled set of target domains.