Study On Algorithms For Microvessel Image Classification Based On Feature Learning

As the pressure of people's work and life is generally increased,the irregular habits of eating and living,the global cancer population has been in a rising trend.Early diagnosis of cancer,is essential to save the lives of cancer patients.Scientific research has found that human microvessel density is associated with the depth of invasion of malignancies and cancer cells.When the length of malignancy exceeds 3cm,new microvessels are needed to provide nutrients and oxygen,while the waste is discharged.Thus,early diagnosis of cancer can be achieved by measuring the density of microvessels.At present,the measurement of microvessels density is mainly dependent on quantitative analysis,that is,the manual counting of the microvessels in hot spots(microvessels distribution of more intensive areas).This approach is not only time-consuming,but also there are differences results between different observers.Modern histopathology uses the algorithms for microvessel image classification to predict cancer,which are still in a very primitive stage.There are many deficiencies,for example,the need of artificial observation,the uncommon use of the classification algorithm.The algorithms for microvessel image classification still need to be optimized and improved.The classification of microvessel image is essentially to divide the microvessel images into two categories,so it can be solved by the classification algorithm in the field of image processing.The main research of this thesis is to study the algorithms for microvessel image classification based on feature learning.The innovation mainly includes the following three parts:(1)Based on the study of the bag of visual words,the fisher vector and the VLAD,the algorithm for microvessel image classification based on spatial features is proposed,including the algorithm for microvessel image classification based on spatial pyramid model,the algorithm for microvessel image classification based on fisher vector and the algorithm for microvessel image classification based on VLAD.Some experiments are carried out to compare and analyze these algorithms.(2)Based on the study of the sparse coding,the algorithm for microvessel image classification based on sparse coding and the algorithm for microvessel image classification based on group sparse coding are proposed.Some experiments are carried out to compare and analyze these algorithms.(3)Based on the study of the deep learning,the algorithms for microvessel image classification based on CNN are proposed.Some experiments are carried out to compare and analyze these algorithms.In this thesis,a variety of algorithms for microvessel image classification based on feature learning are studied.As some experiments are carried out to compare and analyze these algorithms,the algorithm for microvessel image classification based on group sparse coding is proposed.Experiments show that the algorithm for microvessel image classification based on feature learning can effectively solve the classification of microvessel images and provide a reliable reference for early prediction of cancer.