Research On Bone Marrow Cell Images Aided Acute Myeloid Leukemia Diagnosis Method

Leukemia is one of the ten major malignant tumors.Acute myeloid leukemia(AML)has the highest incidence and high mortality among the various types of leukemia.AML originates from bone marrow which is an important hematopoietic organ of human body.Its hematopoietic function is essential for life maintenance.Examining of bone marrow cell morphology and quantitative analysis is helpful for the diagnosis and differentiation of AML.Traditional manual microscopy method seriously delays treatment time because of the long time,and the results are extremely susceptible to personal professionalism and subjectivity.In order to achieve the standardization and standardization of digital pathological diagnosis of bone marrow cell images,we can use computer image analysis and artificial intelligence technology to segment,detect,recognize and count the cells of bone marrow microscopic image.It is of great significance for improving the level and efficiency of clinical diagnosis of AML to overcome the problems that only rely on artificial microscopy.At present,in order to realize the digital assistant diagnosis of AML,it is necessary to solve the key technology of cell location recognition and counting in the image of bone marrow cells.The main problems are as follows: 1)The background area is relatively complex,which makes it difficult to segment and detect individual cells;2)There are many kinds of bone marrow cells are complex,and various types of cells have different shapes but similarities,so it is difficult to distinguish them.This paper focuses on the AML-assisted diagnosis method based on bone marrow cell images,mainly doing the following work:(1)An individual cell detection method based on morphological characteristics is proposed.K-means clustering and regional connectivity judgment are used to segment the nucleus rapidly.Individual cell region judgment and optimization modeling are established for the problem of over-segmentation of the nuclear region.And based on the cytoplasmic search model of regional connectivity,the complete contour of individual cells is inferred and the detection of individual cells is realized.The experimental results show that it has a good individual cell detection effect on granulocytes with complex morphology.(2)Bone marrow cell detection method based on CNN.Cells were detected and classified on the image of bone marrow cells with multiple cells.The anchor based on cluster design is used to predict the target on the multi-scale detection network,and the improved UIoU algorithm alleviates the influence of annotation data errors.In addition,the cascading classification algorithm to classify similar cells improves the classification rate.Experimental results show that the proposed detection method has better localization and classification performance than other methods.(3)Designed and implemented an AML-Aided diagnosis system for bone marrow cell images.The main functions of the system include image selection,cell detection and risk assessment.The main algorithm implemented by the system is based on the CNN-based bone marrow cell detection method proposed in this paper.