Research on Bone Marrow Cell Image Classification Based on Deep Learning
Currently,the diagnosis and treatment of hematopoietic system diseases face numerous challenges,among which precise analysis of the morphology and quantity of bone marrow cells is a crucial step for accurate diagnosis.However,traditional examination methods primarily rely on manual operation,which is constrained by efficiency and accuracy,making it difficult to meet the stringent requirements of modern medical diagnosis.In response to the challenges of bone marrow cell image classification,deep learning tech-nology is employed in this research,with Vision Transformer(ViT)selected as the base model for further research into its classification capabilities.To further optimize the model's classification performance,an innovative convolution-based Input Embedding module is designed.This module cleverly combines the strengths of convolutional neural networks and the attention mechanism,thereby more ef-fectively capturing and encoding key features in bone marrow cell images.By integrating the Input Embedding module,the improved model(Rc-ViT)demonstrates higher accuracy in processing various bone marrow cell images.Experimental results show that compared to the original ViT model,the accuracy of Rc-ViT has increased by 0.024 9,indicating a significant performance enhancement.Espe-cially when dealing with cell categories with limited data,the precision and recall rates have increased by 0.036 0 and 0.150 0,respec-tively.This remarkable improvement not only verifies the effectiveness of this method in enhancing the model's recognition of minority cell categories but also reduces the impact of insufficient data on feature learning.Furthermore,this model has made significant progress in the accuracy and consistency of cell classification,providing strong support for the reliability and repeatability of medical diagnosis.
Deep learningBone marrow cell imagesVision Transformer modelAttention mechanism
NETL
NSTL
万方数据
