CellsBin Presents Novel AI Platform at ASH 2024

Back

This study showcases an innovative approach for the detection and isolation of rare cells utilizing a state-of-the-art transformer-based deep learning model. Our method integrates surface expression data with morphological imaging through a cutting-edge microfluidic system, allowing for the precise discrimination of target rare cells from other blood components without reliance on negative selection. This technological advance addresses the inefficiencies of traditional methods, which typically involve cumbersome enrichment processes and result in low recovery rates and compromised cell viability. The deep learning model was trained using a large data set of cells from a Multiple Myeloma (MM) cell line used for validation purposes and healthy Peripheral Blood Mononuclear Cells (PBMCs) as negative controls. The AI system operates within a dynamic context window, processing consecutive images to jointly track the trajectory of target cells and classify their types. Additionally, the system includes a mechanism to adjust the sensitivity threshold (T), enhancing the balance between detection precision and minimizing false positives. Preliminary testing with cell lines demonstrated the model's effectiveness. For the task of distinguishing MM plasma cells from healthy PBMCs and setting the false-positive to 1 in 3 million, or roughly one false-positive in 1mL of whole blood, the detection rate reached 42.1% (95% CI ± 4.2%). This increased to 79.1% (95% CI ± 6.6%) when the threshold was adjusted to permit two false positives per mL of whole blood. The results highlight the transformative potential of employing advanced AI algorithms in the field of rare-cell biology. Our model's ability to process multi-modal image data significantly improves the detection and isolation of rare cells, such as those associated with MM, and is adaptable for other rare cell types including circulating tumor cells and stem cells from various biological fluids. As we advance our technology, ongoing clinical experiments are focused on refining the assay to ensure robust performance across different stages of disease and in various applications, establishing a versatile tool for real-time monitoring and precision medicine.