A quick prediction for the transformation of red blood cell morphology with deep learning technique

Red blood cells (RBCs) are essential for maintaining human health, and their morphological abnormalities serve as critical diagnostic indicators for various blood disorders. To predict RBC morphology accurately and efficiently, we develop a deep learning model that solves the inverse problem of inferring RBC shapes from their geometric properties. The model utilizes spherical harmonics to reconstruct the RBC surface, thereby reducing the degrees of freedom and enabling efficient prediction. It is applied to predict the stomatocyte–discocyte–echinocyte (SDE) transformation of RBCs, with results strongly correlating with experimental observations and numerical simulations, thereby validating the model's accuracy. Additionally, the model demonstrates robustness with low sensitivity to structural parameters, such as dataset size and hidden layer depth. Even with a small dataset and shallow hidden layers, it remains effective, with training completed in 0.25 h on a standard laptop. Furthermore, we generate a phase diagram of SDE transformation under varying area differences, which further emphasizes the model's accuracy and computational efficiency. This work contributes to a deep learning approach for predicting RBC morphology transformations, offering a computationally efficient and accurate alternative to traditional methods. The model's ability to reproduce RBC morphology by controlling geometric features holds significant implications for understanding and diagnosing blood disorders.