Research On The Method Of AML Cell Image Segmentation And Recognition

Acute myeloid leukemia is a disease of the blood and bone marrow cancer deteriorated rapidly.In AML,the bone marrow produces an excessive amount of immature cells called blast cells.Usually,the blast cells continue to mature within the bone marrow and then begin to carry out set functions.White blood cells constitute the body's immune system,attacking and destroying invasive bacteria and viruses.Red blood cells carry oxygen throughout the body and release carbon dioxide to be exhaled by the lungs.However,excessive blast cells can cause these functions to be not performed properly and effectively.Therefore,in order to realize the diagnosis of AML,the count of the mother cell is the key method.Therefore,in order to realize the diagnosis of AML,the counting of the blast cells is the key method.Traditionally,the manual detection of the blast cells is not only time-consuming,but also easily affected by human subjective attitude and technical level.In addition,the AML cell image itself has many complex characteristics,such as many kinds,different forms and serious overlap,which makes the diagnosis result more difficult to determine.Based on the above analysis,an auxiliary detection method for bone marrow smear is proposed in this paper.The research contents are as follows:First of all,in order to achieve the high accuracy of the aggregation and adhesion region segmentation,a AML cell image automatic segmentation algorithm is proposed in this paper,which implements the two-level segmentation theory based on spatial clustering and hidden Markov random field.the algorithm is based on the color feature of pixels.In the Lab color space,the improved k-means clustering method is used to obtain the initialization tag set.The spatial expression model of the cell image is constructed by HMRF,which makes full use of the spatial constraint relation to reduce the influence of the isolated points and smooth the segmentation area.The model parameters are optimized by using the expectation maximization algorithm.Secondly,on the basis of the two-level segmentation of cell images,we extract the multidimensional feature parameters of all kinds of cells,and use BP neural network classifier to classify and identify all kinds of cells.Neural network has strong learning performance and good expansibility.In this paper,a three-layer network structure is adopted,and ten times cross validation method is used to train the classifier,so as to realize the identification of all kinds of cells.Experiments on about 1800 cells from 61 microscopic images of bone marrow smear showed that the accuracy of the auxiliary system in cell segmentation and cell recognition is over 95%,which has achieved the work of AML disease diagnosis.This method is not only limited to the segmentation and recognition of AML cell images,but also for the diagnosis of other diseases in medicine,such as cervical cancer,lymphatic cancer and so on,and it can also get accurate results.