Bone marrow cell classification and recognition method based on transfer learning and image texture features

Abstract

The invention discloses a bone marrow cell classification and recognition method based on transfer learning and image texture features. The method comprises the following steps: obtaining training sample pictures; extracting cell image textural features and classifying the cell image textural features into large classes; carrying out transfer learning, carrying out machine learning algorithm training on the cell type with the largest number of pictures in the first large class, establishing a model, and carrying out machine learning on the cell type with the second number of pictures in the first large class after the model training is completed; learning a cell type with a third picture number in the cell large class until all types of cells in the cell large class are trained, and then carrying out cell identification training of the next large class; performing algorithm effect determination. According to the method, the defects in the prior art are overcome, and training can be completed with only one tenth of the data volume of a traditional machine learning algorithm; meanwhile, the stability of a machine learning algorithm is improved in combination with traditional textureimage features, so that the problem of recognition and classification of bone marrow cells can be well solved.

Description

technical field [0001] The invention relates to the technical field of medical image processing, in particular to a bone marrow cell classification and recognition method based on migration learning and image texture features. Background technique [0002] Microscopic examination of bone marrow cells is an important basic technology for the diagnosis and differential diagnosis of various diseases including blood diseases, including various types of leukemia, lymphoma, multiple myeloma, anemia and bone marrow metastasis of solid tumors caused by various causes Pancytopenia, parasites and fungal infections, etc. are the primary examination methods for the comprehensive diagnosis of morphology, immunology, cytogenetics and molecular biology (MICM) that are currently recommended for blood diseases. The "Guidelines for the Diagnosis of Myeloid Malignant Tumors" issued by the World Health Organization stipulates that necessary accurate and detailed bone marrow microscopy is requir...

AI Technical Summary

Problems solved by technology

However, due to the variety of bone marrow cells, there are nearly 50 types, and the machine learning of all types of cell shapes requires a huge amount of image data resources, and the efficiency of algorithm training is low; at the same time, the shape similarity of some types of cells is high, and the algorithm is difficult to distinguish Therefore, the actual effect of traditional artificial intelligence methods in the classification and identification of bone marrow cells is not ideal

Images