Study design and results. (IMAGE)
Caption
a, Schematic overview of the study workflow. Blood was sampled three times in the sepsis recovery cohort and once in the healthy control cohort. CD8+ T cells were extracted from the blood using a Magnetic Cell Separator (MACS). 3D cell images were acquired using holotomography. The morphological features of the 3D cell and the 2D sectioned images were obtained and compared. The spatial distribution within the cell was compared with that of the shell structure. Deep learning models for predicting the diagnosis and the prognosis of sepsis were developed and validated based on internal cell structure. PBMCs, peripheral blood mononuclear cells; ICU, intensive care unit. b, Spatial distribution within cells at each time point in longitudinal sepsis recovery and healthy control. Spatial distribution within cells in survival and non-survival groups at the first time point of sepsis recovery (T1) with the density of shell structure. c, The receiver operating characteristic (ROC) curve of the proposed method with one–five cells in predicting diagnosis and prognosis models. d, Validation of the morphological feature through correlation with clinical features and the visual explanation.
Credit
by MinDong Sung, Jong Hyun Kim, Hyun-Seok Min, Sooyoung Jang, JaeSeong Hong, Bo Kyu Choi, JuHye Shin, Kyung Soo Chung,Yu Rang Park
Usage Restrictions
Credit must be given to the creator.
License
CC BY