Label-Free Identification of Lymphocyte Subtypes Using Three-Dimensional Quantitative Phase Imaging and Machine Learning

This method provides an alternative to conventional fluorescence labeling and flow cytometry analysis procedure, which are time-consuming, costly, and incur the risk of altering the cellular function of the samples. The main advantage of this technique is the three-dimensional refractive index tomography and machine learning are label-free and quantitative methods that enable a rapid and accurate lymphocyte identification. The implications of this technique extend toward the therapy of blood cancers and autoimmune diseases as identification of lymphocytes can be crucial for disease diagnosis and appropriate treatment application.

Though this method can provide insight into lymphocyte populations, it can also be applied to the analysis of other single cells of interest, including bacteria. Visual demonstration of 3D quantitative phase imaging technology is critical, as it facilitates a clear instruction of how to perform the technique and can provide insight into its applications. Begin by collecting each lymphocyte subset by fluorescence-activated cell sorting.

For optimal imaging, dilute each cell sample to a concentration of 180 cells per microliter of RPMI medium, and slowly inject 120 microliters of the first diluted sample into an imaging chamber. After confirming a lack of bubbles within the chamber, place a drop of distilled water onto the objective lens of a 3D quantitative phase microscope, and place the imaging chamber onto the translation stage of the microscope. Adjust the stage so that the sample aligns with the objective lens, and click focus and surface in the calibration tab of the microscope perspective of the imaging software to adjust the axial positions of the objective and condenser lenses, respectively.

Click Auto mode to align the objective and condenser lenses. To optimize the alignment, open Scanning mode and manually adjust the lenses to align the digital micromirror device pattern to the center. Then, return to Normal mode and adjust the translation stage to locate a cell in the field of view.

Adjust the axial position of the objective lens to find the focal plane until the sample boundary visualized in the screen is almost invisible. It is important to perfectly adjust the focus of the cell to generate an optimal 3DRI tomogram. If the image is not taken properly, 3D reconstruction will be impaired, resulting in a noisy tomogram.

Adjust the translation stage to find a location without a cell and click Calibrate to measure multiple 2D holograms with varying illumination angles. Adjust the translation stage to locate a cell at the center of the field of view. And under the Acquisition tab, name the sample being imaged.

Click 3D Snapshot to measure the holograms of the cell using the same illumination angles as for the 2D holograms just measured. When the acquired data appears in the Data Management panel, right click the data and click Process to reconstruct a 3D refractive index tomogram from the 2D holograms using the defraction tomography algorithm implemented in the imaging software. After imaging, in the Data Management panel, right click on the data and click Open to visualize the data.

Click the center of the cell to reposition it and click RI Tomogram on the Data Manager panel. On the Preset tab click Load and double click lymphocyte. xml, which is a predefined transfer function provided by the imaging software to visualize the tomogram according to the 3DRI distributions.

Scroll the mouse to zoom in and drag the cell to rotate it any direction. For quantitative morphological and biochemical feature extraction place all of the tomographic data in a single folder and split the cell types within individual subfolders in the main folder. Next, open the supplementary feature extraction file in the appropriate imaging software and edit line 14 to designate the tomogram folder from which the data is to be extracted.

Edit line 15 to designate the folder to which the extracted feature data is to be saved and execute the code. For every tomogram in the dataset the code will calculate the surface area, cellular volume, sphericity, protein density, and dry mass per refractive index threshold. For supervised learning and identification use the simple random splitting algorithm in MATLAB to randomly split the extracted feature data into separate training and test set folders.

Open the supplementary training file and edit line 14 to designate the training set folder, line 16 to designate the folder for saving the trained classifier, and line 17 to set a file name for the classifier. Then execute the code. Using the selected features of the training set, the code will train a classifier with the K nearest neighbor algorithm and save the classifier in the designated folder.

Next, open the supplementary testing file three, and edit lines 14 through 15 to designate the trained classifier to be tested and line 17 to designate the trained test set. Then execute the code. The classifier will identify the cell types of the individual lymphocytes in the test set.

Here representative 3D rendered refractive index tomograms of B lymphocytes, CD4 positive T lymphocytes, and CD8 positive T lymphocytes with different color schemes, allocated according to the refractive index values assigned via the imaging software are shown. From the refractive index values quantitative morphological and biochemical features can be calculated. In this experiment the accuracy of the T and B lymphocyte classification was 93.15%and 89.81%for the training and test cases respectively.

The CD4 positive and CD8 positive T lymphocytes were statistically classified, and the accuracy was 87.41%and 84.38%for the training and test sets respectively. Lastly, the accuracy of the multiclass, cell type classifier was 80.65%and 75.93%for the training and test stages respectively. The quality and number of images are essential to the success of this technique.

The better the quality of image and the higher the volume of data, the better the accuracy of the identification. Deep learning can be used to more completely analyze the complex data of the tomograms, highly enhancing the identification performance. Further, fluorescence and 3D quantitative phase imaging can be used to investigate the path of physiological roles of the identified lymphocytes.

After its development this technique paved the way for researchers in the fields of cell biology and biomedicine to explore specific diseases of interest within different organisms. Indeed, immunologists in particular, may benefit from using this new technology for identifying population of interest.