Fig. 6: Artificial intelligence-based modeling of cell biophysics properties related blood diseases.

a I The finger-press type microfluidic/smartphone imaging system collects the integrated images, morphological, and mechanical parameters loaded into cloud computing. II The pathological diagnosis is performed via cloud computing based on deep learning. III The three fully connected layers contain 40, 64, and 20 vectors, respectively, of which the first fully connected layer contained 32 image vectors and 8 morphological and mechanical parameter vectors. IV Feature extraction of the multiparameter training models in 432 clinical samples. V Diagnostic reports are transmitted to the physician via the cloud sharing platform. b Identification of the primary objective for health, MA, MF, IDA, TTP, Thai. blood samples. c–f Distributions of diameter, circularity, axis ratio, and deformability of the samples. g Confusion matrices comparing the performance on three training pathological diagnosis models between expert diagnosis and deep learning