High speed single-frame label-free cell tomography
A SIngle-frame LAbel-free Cell Tomography (SILACT) system and methods are provided to reconstruct 3D Refractive Index (RI) distribution of cells at over 10,000 volumes/second while resolving subcellular compartments without fluorescence labelling. The SILACT includes a high-speed interference microscope with multiplex illumination and a fast reconstruction method utilizing a pre-trained physics-incorporating Deep Neural Network (DNN). With SILACT, it is demonstrated that 3D imaging cytometry at a throughput of over 20,000 cells/second can be achieved, and transient dynamics of Red Blood Cells (RBCs) undergoing shear-induced 3D deformation inside a microfluidic channel can be observed.
1 . A single-frame label-free cell tomography (SILACT) system for reconstructing three-dimensional (3D) refractive index (RI) value distribution of cell samples, comprising:
an interferometric microscope device with angle-multiplexing optics to capture one or more interferograms of the cell samples; and
a machine-learning (ML) engine configured to reconstruct a 3D RI map of the cell samples based on the one or more interferograms captured;
wherein for each cell sample, the interferometric microscope device is configured to acquire a multiplex single-frame interferogram of the cell sample by scanning the cell sample simultaneously with a plurality of beams having a same elevation angle and different azimuthal angles.
2 . The SILACT system of claim 1 , wherein the ML engine is configured to compensate for missing spatial frequency information.
3 . The SILACT system of claim 1 , wherein for each cell sample, the interferometric microscope device is configured to:
acquire a plurality of sequential interferograms of the cell sample, each by scanning from a single scanning illumination angle.
4 . The SILACT system of claim 3 , wherein the different azimuthal angles include 0°, 90°, 180°, and 270°.
5 . The SILACT system of claim 1 , wherein the ML engine comprises:
an interferogram pre-process unit configured to pre-process the interferogram(s) received from the interferometric microscope device; and
a 3D reconstruction unit receiving outputs of the interferogram pre-process unit and configured to reconstruct the 3D RI map and recover the RI values by linear fitting steps.
6 . The SILACT system of claim 5 , wherein the interferogram pre-process unit is configured to:
pre-process the plurality of sequential interferograms received from the interferometric microscope device to generate a ground truth 3D RI map of the cell sample by physical-model based reconstruction steps; and
pre-process the multiplex single-frame interferogram received from the interferometric microscope device to extract a plurality of Phase Approximants by spatial filtering steps.
7 . The SILACT system of claim 6 , wherein the 3D reconstruction unit is configured to:
train a Deep Neural Network (DNN) based on the ground truth 3D RI map and the plurality of Phase Approximants generated by the interferogram pre-process unit to reconstruct the 3D RI map of the cell samples.
8 . The SILACT system of claim 6 , wherein the physical-model based reconstruction steps are based on Learning Tomography Beam Propagation Method (LT-BPM).
9 . The SILACT system of claim 6 , wherein reliability of the physical-model based reconstruction steps is validated by calibration of calibration cell samples.
10 . The SILACT system of claim 9 , wherein the calibration cell samples include polystyrene beads or 3D printed cell phantoms.
11 . The SILACT system of claim 7 , wherein the DNN model is based on a Learning to Synthesize by DNN (LS-DNN) method.
12 . The SILACT system of claim 7 , wherein the DNN model is trained to minimize Negative Pearson Correlation Coefficient (NPCC) loss between the ground truth 3D RI map and output of the DNN model.
13 . The SILACT system of claim 1 , wherein the cell samples comprise NIH/3T3 cell samples.
14 . The SILACT system of claim 6 , wherein the ground truth 3D RI map comprises 100 layers along a z-direction.
15 . The SILACT system of claim 14 , wherein adjacent layers have an interval of 0.21 μm between them.
16 . The SILACT system of claim 14 , wherein a portion of the cell samples are confined to a small axial dimension with |z|≤8 μm that is equivalent to 80 layers.
17 . The SILACT system of claim 14 , wherein within cell support, most organelles are located at layers |z|≤6 μm, where high frequencies are of highest significance.
18 . A single-frame label-free cell tomography (SILACT) system for reconstructing three-dimensional (3D ) refractive index (RI) value distribution of cell samples, comprising:
an interferometric microscope device with angle-multiplexing optics to capture one or more interferograms of the cell samples; and
a machine-learning (ML) engine configured to reconstruct a 3D RI map of the cell samples based on the one or more interferograms captured;
wherein for each cell sample, the interferometric microscope device is configured to:
acquire a plurality of sequential interferograms of the cell sample, each by scanning from a single scanning illumination angle; and
acquire a multiplex single-frame interferogram of the cell sample by scanning the cell sample simultaneously with a plurality of beams having a same elevation angle and different azimuthal angles.
19 . A single-frame label-free cell tomography (SILACT) system for reconstructing three-dimensional (3D ) refractive index (RI) value distribution of cell samples, comprising:
an interferometric microscope device with angle-multiplexing optics to capture one or more interferograms of the cell samples; and
a machine-learning (ML) engine configured to reconstruct a 3D RI map of the cell samples based on the one or more interferograms captured;
wherein the ML engine comprises:
an interferogram pre-process unit configured to pre-process the interferogram(s) received from the interferometric microscope device; and
a 3D reconstruction unit receiving outputs of the interferogram pre-process unit and configured to reconstruct the 3D RI map and recover the RI values by linear fitting steps,
wherein the interferogram pre-process unit is configured to:
pre-process the plurality of sequential interferograms received from the interferometric microscope device to generate a ground truth 3D RI map of the cell sample by physical-model based reconstruction steps; and
pre-process the multiplex single-frame interferogram received from the interferometric microscope device to extract a plurality of Phase Approximants by spatial filtering steps.