A Study On Visual Perception In Mammography Diagnosis

Breast cancer is one of the most common malignant tumors that hazard the middle-aged women's health. In China, the incidence of breast cancer in women presents persistent high growth. Currently, the annual growth rate sharply rises by 3% ~ 4%. Due to the easy, low-cost, non-invasive and good detection results of early asymptomatic of occult breast cancer, mammography becomes the primary method for breast cancer detection in the current clinical environment.CAD technology in mammography can aid the radiologist to diagnosis. Although much progress has been made, there still exist deficiencies in accuracy of mass detection and classification. High false-positive rate hinders its clinical use. How to use the latest information technology to improve the CAD technology especially the tumor detection accuracy, therefore, becomes an important research topic of application value.There are some close intrinsic links between the visual perception behavior of radiologist and the diagnosis progress of medical image. Analysis of the visual perception data in diagnosis progress has important value to improve the diagnostic performance of CAD system. This dissertation respectively views from selective visual attention model and analysis of eye movement information to verify the differences and the resulting perception difference between benign and malignant tumor. The main purposes to achieve are: (1) consistency measurement of points and regions in position between the visual attention computational model and real interest points of radiologist in diagnostic acts; (2) verify the effectiveness of the calculation model in automatically extraction of local features; (3) analysis the statistical differences of radiologist's visual perception information distribution between benign and malignant tumor diagnosis. All of these can promote the later work such as automatic analysis of benign and malignant tumor, modeling of diagnosis pattern.Here, we start from the eye-movement perception information in clinical diagnosis of radiologist to introduce our work in two parts as follows:One is tumor classification which is based on the analysis of visual perception information. After the pretreatment of collected eye-movement data of radiologists' diagnosis of benign and malignant tumors, using a density-based clustering approach to achieve the capture of interest points. Through the calculation of region of interest, selection and extraction of local feature in these regions, the local feature is used for classifier training and testing to prove the effectiveness of local features in tumor classification. In addition, selective visual attention model is used to guide the local feature extraction. The same classifier and classifier parameters is adopted for training and testing. Effectiveness and differences of the two methods in tumor classification is then evaluated objectively according to the classification rate.The other is the statistical analysis of visual perception information. In general, there exist differences in eye-movement perception information, which is performed in the radiologists'viewing of different locations. Through the statistics of Eye-movement perception information, using statistical analysis methods such as descriptive statistics, parametric and non-parametric tests to confirmed the existing of radiologists'perception difference in benign and malignant diagnosis. With these statistics conclusion, we can make the CAD system more intelligent and build the solid theoretical and experimental fundament in exploring the correspondence relationship between eye-gazing and diagnosis decision-making or knowledge set. It has some value in both research and application.Finally, give the summaries and prospects of the full text. In summing up the work of this dissertation and prospecting the direction of future research.