Blood sample image intelligent identification and classification system based on computer vision

By separating the nuclear and cytosolic channels of blood smear images using the Lambert-Beer law and nonlinear concentration mapping, a simulated single-cell layer is constructed and differential calculations are performed. This solves the problem of cell overlap and occlusion in high-density smears, enabling accurate identification and classification of blood cells.

CN122156756APending Publication Date: 2026-06-05JINHUA MUNICIPAL CENT HOSPITAL
0 Cites 0 Cited by

Patent Information

Application Number
CN202610259941.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-03-04
Publication Date
2026-06-05

AI Technical Summary

Technical Problem

Existing technologies suffer from low recognition accuracy when processing high-density blood smears due to cell overlap and occlusion, making it difficult to accurately reconstruct biological characteristics from complex backgrounds and affecting diagnostic accuracy.

Method used

By using illumination correction and nonlinear concentration mapping based on the Lambert-Beer law, cell nuclear channel and mass channel data are separated, a simulated single-cell layer is constructed, and independent single-cell layers are generated through difference calculation and inverse correction. These layers are then combined with a convolutional neural network for classification.

Benefits of technology

It effectively restores the true geometric shape and material distribution of the obscured target, reduces the false positive rate, and achieves refined classification and high-accuracy diagnosis of blood cells.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention discloses an intelligent blood sample image recognition and classification system based on computer vision, and relates to the technical field of pathological image analysis and computer vision, and the system comprises the steps: obtaining an original collection image, executing nonlinear concentration mapping, and generating a biochemical layered image containing actually measured concentration data; calculating and locking a cell center through center response, dividing an initial jurisdiction, and constructing an ideal simulated single cell layer; calculating a difference graph of the simulation superposition image and the actual measurement image, and carrying out iteration reverse correction on the simulation image layer by utilizing concentration difference data until an independent single-cell image layer containing recovered textures is obtained through convergence; and finally, generating an overlapping complexity weight graph based on the difference graph, constructing a multi-dimensional feature data packet, inputting the multi-dimensional feature data packet into the convolutional neural network, and outputting a classification result. According to the method, through combination of physical model decoupling and deep learning, the identification problem caused by cell overlapping in a high-density sample is effectively solved.
Need to check novelty before this filing date? Find Prior Art