Heterotypic lymphocyte typing method based on YOLOv5 and microscopic hyperspectral image

Abstract

The invention provides a heterotypic lymphocyte typing method based on YOLOv5 and a microscopic hyperspectral image. The heterotypic lymphocyte typing method is improved on the basis of a standard YOLOv5 network structure. The method comprises the following steps: synthesizing a lymphocyte pseudo-color image by using gray images corresponding to 461, 548 and 698 nm, and finishing automatic positioning of lymphocytes by facing synthesized image information through an improved YOLOv5 network; using the relative position information of a lymphocyte prediction frame in an improved YOLOv5 network identification result to obtain an average spectrum value, using the average spectrum value as input, and using a one-dimensional convolutional neural network to realize typing identification of the heterotypic lymphocytes; and constructing the heterotypic lymphocyte typing model. By improving the standard YOLOv5 network structure, the network calculation amount is reduced, the accuracy of small target positioning is improved, the idea of first positioning and then recognition is put forward, the detection efficiency of lymphocytes is improved, and typing recognition of heterotypic lymphocytes is achieved.

Description

technical field [0001] The invention belongs to the field of medical image processing, and relates to a method for typing atypical lymphocytes based on YOLOv5 and microscopic hyperspectral images. Background technique [0002] Lymphocytes are blood cells that play an immune role in the body. Under the stimulation of virus, protozoan infection, drug reaction, stress state or allergen and other factors, lymphocytes may undergo cell proliferation and morphological changes such as cell body enlargement, cytoplasm increase, and basophilic enhancement. Atypical lymphocytes are lymphocytes with cell proliferation and morphological changes. Atypical lymphocytes are divided into three types (type Ⅰ, type Ⅱ and type Ⅲ), and the increase or decrease of each type of atypical lymphocytes may represent the occurrence of a disease. In clinical testing, the diagnosis of infectious mononucleosis, viral hepatitis and allergic diseases is mainly realized through the typing and identification...

AI Technical Summary

Problems solved by technology

However, microscopic hyperspectral images contain hundreds of bands, and the huge amount of data has a lot of redundancy, which will cause a disaster of dimensionality to the model.

Images