Blast Cell Image Classification Without CD Marker Panels

The patent segments the classification task into two distinct neural network models: a first model that identifies abnormal promyelocytes and other mononuclear blood cells, and a second model that specifically distinguishes myeloid from lymphoid lineage. This segmentation allows each model to specialize in a specific classification aspect, improving overall accuracy while enabling parallel processing that reduces total diagnosis time compared to sequential CD marker assessments.

The patent replaces the mechanical and chemical CD marker assessment process with a computational image analysis system using deep neural networks. Instead of physically isolating and chemically analyzing cell surface markers, the system uses digital imaging and automated neural network classification to identify and differentiate blast cells, dramatically reducing both time and cost while maintaining or improving classification accuracy.

The patent uses digital copying of blood smear images to replace physical handling and multiple rounds of CD marker analysis. By creating and analyzing digital copies of cell images through neural networks, the system eliminates the need for repeated physical sample processing and chemical reagent consumption, significantly reducing diagnostic costs while maintaining classification accuracy.

The patent employs a cost-effective digital imaging approach that replaces expensive and time-consuming CD marker panels. The neural network models process standard blood smear images without requiring additional expensive reagents or specialized equipment, making the diagnostic process more affordable and accessible.