Feature Analysis And Recognition Of Pathologic Cell Images

Feature quantification and identification based on mico-cell images is one of the mostimportant points for biomedical image analysis, computer assistant/automatic diagnose systemsand clinical applications. The computer assistant/automatic diagnose system combined with theexperts’ experience is aimed to characterize objects by quantified parameters rather then images,also aimed to achieve primary results of classification and identification for final diagnosis. Thesystem is not only beneficial to improving the diagnostic efficiency and saving labor costs, butalso beneficial to avoiding miss abnormal diagnosis caused by the lack of experience and thecomplex tasks.The main contents of this thesis are quantification and automatic classification based oncervical cell images combining the technique of image processing and analysis with the theory ofcytology and experts’ experience. Dealing with the complex and time-consuming task of theartificial cervical lesions census, this thesis proposes an automatic method to fast detectabnormal cells based on feature quantification and clinical practice experience. Dealing with theclusters in cervical pathologic cell images, this thesis proposes methods to extract the featuresand count cell numbers at the same time. Ultimately, a whole plan aimed to classify and identifyfor cervical pathologic cell images based on features quantification is proposed, and a GraphicalUser Interface System is designed. The main contents are as follows:1. Study the status quo of cell image processing, analysis, and cervical cytology from theliteratures in recent years. For the needs of further image analysis, we use the methods based onmathematical morphology to deal with image pre-processing problems, firstly.2. Features quantification is based on the cervical cytology grading standards, such as shape,color, texture, and cyto-pathological characteristics are extracted for single cervical celldescription. An automatic method for fast distinguishing abnormal cells from normal cells isproposed based on feature quantification and clinical practice experience to deal with thecomplex and time-consuming task of the artificial cervical lesions census. 3. Dealing with the clusters in cervical pathologic cell images, this thesis proposes methodsto extract the features for cluster description and count cell numbers at the same time. Theclusters classification is based on the extracted features.4. This article thesis summarizes a complete set of methods for cervical cyto-pathologicalimage features analysis, classification and identification, which combines the cytologicalknowledge and experience in expert diagnosis. A Graphical User Interface System is designedstep by step to primary diagnose the cervical cyto-pathological image into four grades: Normal,Low-grade Squanous Intraepithelial Lesion, High-grade Squanous Intraepithelial Lesion, andSquanous Cell Carcinoma.