Research On Key Techniques Of Automatic Examination Of Bone Marrow Cell Morphology For Acute Leukemia

Acute leukemia is a highly heterogeneous tumor and the incidence rate is the highest among malignant tumors.It seriously threatens the physical health and daily life of patients.Therefore,early diagnosis and treatment are very important for patients with acute leukemia.Morphological examination of bone marrow cells is the earliest application of a blood tumor classification diagnosis method and it is also one of the most basic and important methods for acute leukemia diagnosis.However,nowadays,the traditional method of microscopic examination still adopts the way of manual observation and counting.With the increase of the incidence of acute leukemia,the challenges and problems existing in the morphological examination of bone marrow cells include: the daily inspection workload is large,the manpower is wasted,and the efficiency is difficult to improve;the requirements of medical experts are high,and the recognition results depend on the human subjective judgments;objective quantitative standards in the description of cell morphology lack etc.Therefore,the use of image processing and artificial intelligence technology to achieve objective,automatic bone marrow cell identification and classification and statistics is of great significance for improving the efficiency and level of clinical diagnosis of diseases.At this stage,there are the following problems in automatic system for bone marrow cell morphology examination: 1)A comprehensive and standardized large-scale database of bone marrow cell image morphology is lack,which has limited the development of this field.2)Clinically,the microscopic images of bone marrow smears are complex and changeable.Besides,the individual's fluctuation is large,and the environment of dyeing is fluctuant.At the same time,the bone marrow cells in the bone marrow smear microscopic images are of a complex category,and the shapes of various types of cells are irregular.There are various forms of light-colored cytoplast around the cell,and there are large similarities between the bone marrow cells in the adjacent growth stage,and the background area is also more complicated.Therefore,whether the bone marrow cells are segmented or classified,it is extremely challenging.Therefore,in view of the above problems,this paper has done the following work to realize the acute leukemia diagnosis process based on the automated examination of bone marrow cell morphology.The details are as follows:1.Collect bone marrow cell smear images under a microscope and construct a bone marrow smear microscopic image database and a single bone marrow cell image database.The bone marrow smear microscopy image is directly derived from clinical data covering almost all types of acute leukemia and the different condition of the individual.While the single bone marrow cell image is an image containing only one bone marrow cell,and the image is extracted from the bone marrow smear microscopic image.It contains all types of bone marrow cells,and has the largest variety and the largest amount of data.2.In the face of complicated and changeable bone marrow cells,human error,equipment differences,and a large number of cellular agglomerations and adhesions when hyperplasia is active,we proposed a bone marrow cell segmentation algorithm based on sparse representation and mathematical morphological operation to achieve the segmentation of the single bone marrow cell from clinical marrow smear microscopic images.3.According to the classification characteristics of bone marrow cell morphology,a bone marrow cell classification algorithm based on hierarchical network was proposed.This method used the second-level deep learning structure to autonomously study the classification features of images,and characteristic parameters of bone marrow cells in different categories and growth stages.As a result,the classification recognition rate has improved largely.