Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2022 Oct 15.
Published in final edited form as: Biosens Bioelectron. 2021 Jun 12;190:113403. doi: 10.1016/j.bios.2021.113403

Raman and quantitative phase imaging allow morpho-molecular recognition of malignancy and stages of B-cell acute lymphoblastic leukemia

Santosh Kumar Paidi 1,^, Piyush Raj 1,^, Rosalie Bordett 2, Chi Zhang 1, Sukrut Hemant Karandikar 3, Rishikesh Pandey 2,4,*, Ishan Barman 1,5,6,*
PMCID: PMC8492164  NIHMSID: NIHMS1714887  PMID: 34130086

Abstract

Acute lymphoblastic leukemia (ALL) is one of the most common malignancies that account for nearly one-third of all pediatric cancers. The current diagnostic assays are time-consuming, labor-intensive, and require expensive reagents. Here, we report a label-free approach featuring diffraction phase imaging and Raman microscopy that can retrieve both morphological and molecular attributes for label-free optical phenotyping of individual B cells. By investigating leukemia cell lines of early and late stages along with the healthy B cells, we show that phase images can capture subtle morphological differences among the healthy, early, and late stages of leukemic cells. By exploiting its biomolecular specificity, we demonstrate that Raman microscopy is capable of accurately identifying not only different stages of leukemia cells but also individual cell lines at each stage. Overall, our study provides a rationale for employing this hybrid modality to screen leukemia cells using the widefield QPI and using Raman microscopy for accurate differentiation of early and late-stage phenotypes. This contrast-free and rapid diagnostic tool exhibits great promise for clinical diagnosis and staging of leukemia in the near future.

Keywords: Raman spectroscopy, quantitative phase imaging, label-free microscopy, B-ALL, leukemia, random forest classification, phenotype recognition

1. Introduction

Acute lymphoblastic leukemia (ALL) is one of the most common cancers in the pediatric population due to the malignant clonal proliferation of lymphoid progenitor cells. B-cell ALL (B-ALL) is the leading cause of cancer-related deaths in children and adolescents (Woo et al., 2014). The clinical diagnosis of B-ALL currently relies on the use of multi-parametric flow cytometry for the identification and classification of B-ALL. However, flow cytometry-based assays require expensive high-quality antibodies and extensive sample preparation; additionally, non-specific signals due to antibody cross-reactivity and photobleaching pose challenges. Several other methods such as those based on mass cytometry (Good et al., 2018), DNA methylation (Nordlund et al., 2015), and gene expression (Ross et al., 2003) are also characterized by some of these limitations. Therefore, there is an unmet need for analytical tools that can provide a facile route for quantitative early diagnosis and staging of B-ALL.

In this milieu, techniques based on label-free optical imaging and molecular spectroscopy are emerging as attractive alternatives for cancer cell phenotyping based on their morphological and biomolecular information (Lam et al., 2019, Ahmad et al., 2018, Winnard Jr et al., 2017, Basu et al., 2014, Di et al., 2014, Chhetri et al., 2012, Boustany et al., 2010). Specifically, quantitative phase imaging (QPI) has been used to investigate the morphology of live cells and identify disease-specific cellular phenotypes (Karandikar et al., 2019, Pavillon et al., 2018, Park et al., 2018, Jo et al., 2018, Kasprowicz et al., 2017, Kim et al., 2017, Lee et al., 2017, Hosseini et al., 2016, Kim, Kim et al., 2013, Kim, Yoon et al., 2013, Popescu et al., 2008). On the other hand, Raman spectroscopy has been employed to measure the biomolecular composition with a single-cell resolution for cell differentiation and disease identification (Paidi et al., 2020a, Paidi et al., 2020b, Bergholt et al., 2019, Kumamoto et al., 2019, Schie et al., 2018, Ember et al., 2017, Managò et al., 2016, El-Mashtoly et al., 2015, Okada et al., 2012, Hamada et al., 2008). Since both the molecular and morphological changes are associated with malignant transformations, we have sought to develop morpho-molecular microscopy (3M) platform, which combines the complementary morphological and biochemical information content into diagnostic frameworks (Paidi et al., 2021, Pandey et al., 2019). There are two main motivations to combine Raman spectroscopy and QPI. First, the increased information content of this hybrid modality confers diagnostic advantage by accounting for alterations in both the cellular attributes due to disease onset and progression. Second, the overall sampling time of a Raman measurement can be reduced by imaging only the suspicious cell populations by leveraging high-throughput characteristics of QPI for coarse cellular analysis or screening. In this study, we employed our custom-built 3M system that leverages the widefield imaging capability of QPI and exquisite molecular specificity of Raman spectroscopy to derive complementary data from a panel of B-ALL cells at different stages (Fig. 1) (Pandey et al., 2019). Using visually aided morpho-phenotyping recognition (VAMPIRE) tool, we reveal that the progression of B-ALL is characterized by robust shape mode changes captured by QPI (Phillip et al., 2021, Wu et al., 2015). We also show that the Raman spectroscopic data provided adequate clustering capability based on the cell type in principal component analysis-derived component space. Furthermore, the supervised classification routine based on random forest analysis enabled accurate classification of the B-ALL cells based on both the stage of disease progression and the cell line identity. Such an approach, fusing the biomolecular analysis afforded by Raman spectroscopy with morphological assessment capability of rapid widefield QPI, could be exploited to accelerate the phenotype recognition of single cells by sampling/screening a subset of cells based on their biophysical properties for downstream biomolecular characterization.

Figure 1. Study overview.

Figure 1.

Open in a new tab

Morpho-molecular assessment of B-ALL using Raman spectroscopy and quantitative phase imaging.

2. Material and methods

Healthy control B cells

The fresh blood samples from three anonymous healthy adult donors were purchased from StemCell Technologies (Vancouver, Canada) and all the experiments were conducted within 24 hours of blood donation. The purchased blood samples contained ethylenediaminetetraacetic acid (EDTA) as an anticoagulant. The B cells were isolated from fresh blood samples using negative selection kits from Stemcell Technologies (EasySep Direct Human B Cell Isolation Kit). This separation kit uses immunomagnetic negative selection for isolating B cells from the whole blood sample. The isolation was carried out following the manufacturer’s instructions and B cells were isolated in phosphate-buffered saline (PBS) free from Ca++ and Mg++ (Gibco, Thermo Fisher Scientific). The B cells suspended in the PBS from the final incubation were centrifuged at 400g for 5 minutes. The cell pellet was resuspended in PBS before the cells were imaged.

Cell lines

We purchased the early-stage leukemia cell lines REH and RS4;11 from American Type Culture Collection (ATCC, USA) and procured the late-stage leukemia cell lines BALL-1 and MN60, from RIKEN BioResource Research Center (Japan) and DSMZ (Germany), respectively. All four cell lines were cultured according to previously described protocols before imaging measurements (Ayyappan et al., 2020).

Flow cytometry

Both primary cells and cell lines were tested using multiparametric flow cytometry for their purity.

Anti-CD10-PerCP/Cy5.5 (Cat no. 312215), anti-CD19-APC (Cat no.392503), anti-CD20-FITC (Cat no. 302303), and anti-CD38-PE (Cat no 356603) were purchased from Biolegend. Fluorescent viability dye, eFlour780, was purchased from ThermoFisher Scientific (Cat no. 65–0865-14). Cells were stained with individual antibodies (quantity as recommended by the manufacturer) in cold PBS containing 0.5% fetal calf serum and the fluorescent viability dye (1:1000 dilution) for 20 minutes. Cells were maintained at a density of one million cells per 100 ml of buffer. Cells were then washed three times with 200 μl of the buffer and resuspended at the above density. Control (unstained) cells were stained with just the fluorescent viability dye. Flow cytometry was performed to analyze the stained and control cells on a MACSQuant Analyzer, and the acquired data were further analyzed and plotted using FlowJo. At least 30,000 cells were acquired from each sample.

Quantitative phase imaging and analysis

We acquired quantitative phase images of cells by suspending 10–15 μL of cell suspension from cultures at a density of 1×106 cells/mL sandwiched between coverslips separated by a secure seal spacer (Invitrogen). We employed diffraction phase microscopy of our 3M system described previously for acquiring interferograms (Pandey et al., 2019). The raw interferograms were processed using the inverse Fourier transform method to reveal phase images. We performed cell segmentation for confining only one cell in a frame for quantitative analysis and obtained 2294 images from the cells in all five groups. We performed QPI measurements from three replicates of the same culture and also from the three different cultures of each cell type. The B cells were measured from three different donors. To understand the differences in the morphological differences between the cell types, we employed the recently developed VAMPIRE tool to analyze the QPI images (Phillip et al., 2021, Wu et al., 2015).

Raman spectroscopy and analysis

We employed the 3M system, described previously, with 785 nm excitation for the acquisition of Raman spectra (Pandey et al., 2019). The cell samples were centrifuged and resuspended in media in PBS before transferring them to quartz cover glass for Raman measurements. For the cell lines, around twenty measurements were performed from each cell culture with three replicas to account for the variation in different measurements from the same culture. Moreover, cells were also measured from three different cultures. Similarly, the spectra from healthy B cells were recorded from cells from three different donors. We collected Raman spectra from cells in the Control (n=123), REH (n=198), RS4;11 (n=200), MN60 (n=202), and BALL-1 (n=236) classes. The integration time for each Raman spectrum (two acquisitions at 30s exposure) was 60 seconds. All the Raman data analyses were performed in MATLAB environment (Mathworks) unless otherwise noted. The spectra in the fingerprint region (700–1750 cm−1) were subjected to background removal using an iterative fifth-order polynomial subtraction method (Lieber and Mahadevan-Jansen, 2011). The background-corrected spectra were subjected to peak normalization using the maximum intensity of each spectrum. We used principal component analysis for dimensionality reduction of the spectral dataset and identify a set of orthogonal abstract component spectra ordered according to the variance in the dataset they explain (Jolliffe, 2002). We used Orange data mining software to plot the scores of the principal components to visualize the clustering of the spectral data in the principal component space according to their cell type and their stage in leukemia progression (Demšar et al., 2013). We used TreeBagger class in MATLAB to implement Breiman’s original random forest algorithm for supervised classification of the spectral dataset (Breiman, 2001). We included 100 trees in each random forest and performed 200 training iterations by selecting random subsets of spectral datasets in each iteration to ensure the equal size of all the classes in the trained model and avoid bias due to unequal class sizes. The out-of-bag classification error rates were calculated for each trained model by subjecting the spectra that are intrinsically left out as part of random selection of examples during training the classifier.

3. Results and discussion

We used leukemic B cells from early (REH, RS4;11) and late (BALL-1, MN60) stages of malignancy along with healthy B-cells (Control) for morpho-molecular characterization using QPI and Raman spectroscopy. Figure 2A illustrates the expression of various known surface markers to assess the purity of the cell lines. We also determined the percentage enrichment of healthy B cells from the isolated B cells (Fig. 2B). Next, we acquired QPI images of cells from REH (n = 389), RS4;11 (n = 406), BALL-1 (n = 394), MN60 (n = 415), and Control (n = 690) classes (Fig. 3A). To characterize morphological changes associated with B-ALL progression, we employed the recently developed VAMPIRE tool to analyze QPI images (Phillip et al., 2021, Wu et al., 2015). The VAMPIRE tool identifies differences in morphological phenotype in terms of the distribution of the cell population in each class into few representative shape modes determined by dimensionality reduction of morphological parameters (Phillip et al., 2021, Wu et al., 2015). By subjecting the entire QPI dataset comprising of 2294 images to the VAMPIRE analysis, we determined five unique shape modes that capture variance in cellular morphology across the dataset (Fig. 3B). The distribution of cells into these shape modes was determined by K-means clustering to identify the representative shape mode for each cell in the dataset. We observed distinct patterns of the cell shape mode distributions for the cells that are indicative of the differences in the stages of B-ALL progression. While the distribution for the Control group is skewed towards shape modes 1 and 2 and the distributions for the late-stage cells are skewed towards shape modes 4 and 5, the distributions for early-stage cell groups are centered around the middle shape modes marking a transition between the healthy and late-stage cells (Fig. 3C). While the shape modes constitute abstract transformations of conventional morphological descriptors, analysis of these shape modes clearly reveals that there exist characteristics of B-ALL progression that are directly related to the morphological attributes of these cell populations. Furthermore, the robustness of such abstract markers based on cell morphology is confirmed by the similarity of the distributions for both the cell lines within each stage-specific group of B-ALL progression. On the other hand, this similarity makes the discrimination between cell types at similar stages based solely on morphological parameters challenging.

Figure 2. Flow cytometry characterization of B-ALL cell lines and control healthy B-cells.

Figure 2.

Open in a new tab

(A) The expression profiles of characteristic surface markers of B-cells are shown for the cell lines used in the study. The flow cytometry data shows the fluorescence intensity vs. normalized counts to show the distribution of the stained and unstained cells for each marker. (B) The percentage enrichment of healthy B-cells from the isolated B-cells as assessed using flow cytometry is shown.

Figure 3. Quantitative phase imaging of B-ALL.

Figure 3.

Open in a new tab

(A) Representative phase images of the Control, early-stage (REH and RS4;11), and late-stage (BALL-1 and MN60) B-ALL cells are shown. (B) The histograms show the relative abundance profiles of the five shape modes identified using VAMPIRE analysis in each cell group. The overlapping curves fitted to the histogram show the skew of the distribution of the shape modes in each cell group. (C) The comparison of the fitted curves shows the progression of the abundant shape mode from right to left in response to the malignant progression of B-ALL.

To tackle this limitation of morphological analysis, we employed Raman spectroscopy to complement the morphological information captured by QPI with rich molecular information indicative of the chemical makeup of the B-ALL cells. We used our custom-built Raman microscope to collect a total of 959 Raman spectra from cells in Control, REH, RS4;11, BALL-1, and MN60 classes. The mean and standard deviation of the spectra in each class are shown in Fig. 4A. The spectra in the fingerprint region exhibit peaks representative of biological components of cells, particularly at 790 cm−1 (O-P-O stretching of DNA), 1008 cm−1 (C–C stretching vibration of phenylalanine), 1125 cm−1 (C-C stretching mode of lipids and C-N stretch of proteins), 1452 cm−1 (CH2 bending modes in lipids and proteins) and 1654 cm−1 (amide I of proteins and C=C stretching in lipids) (Movasaghi et al., 2007). We performed principal component analysis of the dataset to perform dimensionality reduction and obtain a set of components ordered according to the variance in the spectral dataset they capture (Fig. S1, Supporting Information). Each spectrum in the dataset is described in terms of these component loading vectors and the corresponding scores. We used the scores of the PC loadings to investigate if specific clustering patterns exist in the spectral dataset that correlate with the stage of the disease progression. Using the Radviz radial visualization tool in Orange data mining software, we plotted the scores of the component loadings that maximize the separation of the clusters based on the cell type (Fig. 4B) (Demšar et al., 2013). We observed appreciable clustering of spectra from each cell type and localization of the clusters belonging to cells in the same stage of progression on the Radviz score map. The larger overlap of the early-stage REH and RS4;11 clusters in comparison to the late-stage BALL-1 and MN60 clusters hints at the divergence in molecular composition between different leukemia cells stemming from further disease progression. However, the reasonable separation of both early-stage and late-stage cell clusters from the Control cluster indicates that Raman spectroscopy can capture subtle molecular changes associated with leukemia – even during the early stages of transformation.

Figure 4. Raman spectroscopy of B-ALL.

Figure 4.

Open in a new tab

(A) Shaded error plots show the mean and standard deviation of the Raman spectra obtained from cells in each group. The key spectral features of the spectra are highlighted. (B) The radial visualization plot shows the clustering of the Raman spectra obtained from the Control and B-ALL cells in the principal component space.

Motivated by these promising results, we sought to employ supervised classification techniques for quantitative assessment of Raman spectroscopy for identification of the leukemia cells. Therefore, we employed random forests – a method based on ensemble of decision trees – for developing classifiers for the present application. We performed multiclass random forest analysis by using the identity of the five cell lines as the class labels for the spectra. Specifically, we trained five-class random forest classifiers composed of 100 decision trees and computed the out-of-bag classification error, which is a measure of performance based on testing the classifiers using samples that have been left out of training. To further improve the robustness of the classification results, we performed 200 iterations of random forest classification by including random subsets of spectra from each class to ensure equal representation of all the classes in the model training step. We observed that the out-of-bag classification error decreased below 5% through the inclusion of ~65 trees for the five-class classification (Fig. 5A). We also found that the high accuracy of classification is consistent for all the five cell lines (Fig. 5B). While the high classification accuracy levels underscore that the acquired Raman spectra encode subtle, yet reproducible, differences in the biomolecular composition of various malignant leukemia cell types, it is also important to verify that the developed classifiers can be used to identify the stage of disease progression. Therefore, we performed the same random forest analysis by using the stage of leukemia progression as an outcome to train a three-class model for the classification of control, early-stage, and late-stage cells. The out-of-bag error rate for the three-class classification based on the stage of the cells in leukemia progression also showed a similar decline to below 5% by including only 65 trees (Fig. 5C). The accuracy for recognition of individual classes also shows that supervised classification based on Raman spectra can accurately distinguish healthy B-cells from early-stage and late-stage leukemia cells (Fig. 5D). It is interesting to note that the out-of-bag error rate for the three-class classification (Fig. 5C) exhibits a relatively higher standard deviation across the iterations, compared to the five-class classification (Fig. 5A). We attribute the higher standard deviation to the heterogeneity introduced within each class of the three-class classification models due to the pooling of two different cell lines at similar stages of leukemia progression. Such heterogeneity is expected to make the classification slightly challenging compared to the five-class model where each class comprises of cells from the same cell line. The confusion matrices for the five-class (Fig. 5E) and three-class classification (Fig. 5F) models show high accuracy for each class and that the few misclassifications occur into the phenotypically similar neighboring classes.

Figure 5. Random forest classification of Raman spectra.

Figure 5.

Open in a new tab

(A) The out-of-bag classification error plot, which illustrates the evolution of classification error as a function of the number of decision trees included in the forest, shows that the error for random forest classifiers trained on the Raman spectra from the five cell lines falls below 5% with the inclusion of fewer than 100 trees in the forest. The shadow represents the standard deviation of error across 200 iterations. (B) The corresponding classification accuracy for each B-ALL cell line and healthy B-cells are plotted for all the 200 iterations. (C) The out-of-bag classification error plot shows the evolution of error for the random forest classifiers trained on the stage of the cells in B-ALL progression. (D) The corresponding accuracy for control, early-stage, and late-stage cells are shown for 200 iterations. The confusion matrices for the (E) five-class and (F) three-class classification models show the percentage of class-wise predictions obtained in the 200 iterations.

While these results showcase the promise of this versatile and widely accessible morpho-molecular platform, it is worth noting the limitations of the present proof-of-principle study. The QPI and Raman measurements have not been performed on the same set of cells thereby limiting our ability to infer the direct benefit of fusing these two datasets. Nevertheless, the acquisition of a significantly large number of images and spectra accounts for the cellular heterogeneity to a large extent through an adequate sampling of the populations. Furthermore, the recent improvement in our system now enables sequential Raman and QPI measurements from the same cells allowing us to directly address this drawback in our future studies. Overall, our results show that while morphological characterization of the leukemia cells provides a clear demarcation between the cells at various stages of disease progression, the rich molecular information captured by Raman spectroscopy can provide a route for additional demarcation between different cell groups at similar stages.

4. Conclusions

Our proposed approach makes it possible to distinguish between healthy, early-stage, and late-stage cells B cells and show the promise of this morpho-molecular platform for fully automated, label-free, and rapid diagnosis of B-ALL. Importantly, a recent study from our laboratory indicates that QPI can be employed to discriminate B-cells from other leucocytes (Shu et al., 2020). This would obviate the need for any additional separation method to select B-cells either from the whole blood or from the leucocyte mixtures. Consequently, the leucocyte mixture, which can be very easily separated from the peripheral blood using centrifugation, can be directly used for the proposed hybrid assay. This fully label-free strategy provides a unique platform in which the malignant B-cells can be optically interrogated while also preserving them for subsequent cytogenetic and immunophenotypic analysis.

Supplementary Material

1

Highlights.

  • Label-free morpho-molecular imaging, in combination with machine learning, allows identification and staging of B-ALL cells.

  • Quantitative phase imaging captures subtle morphological differences consistent with the stage of malignant progression of B-ALL cells.

  • Raman spectroscopy allows prediction of not only the stage of malignant progression but also different cell types within each stage.

Acknowledgments

This work was partially supported by Connecticut Children’s, National Cancer Institute (R01 CA238025), the National Institute of Biomedical Imaging and Bioengineering (2-P41-EB015871-31), and the National Institute of General Medical Sciences (DP2GM128198). S.K.P. acknowledges the support of the SLAS Graduate Education Fellowship Grant. We also acknowledge the support from the Carole and Ray Neag Comprehensive Cancer Center, University of Connecticut School of Medicine. The schematic in Figure 1 was partially created with BioRender.com.

Footnotes

The authors disclose no potential conflicts of interest.

CRediT author statement

Santosh Paidi: Investigation, Methodology, Formal analysis, Visualization, Writing - original draft.

Piyush Raj: Investigation, Methodology, Formal analysis, Writing - review and editing, Data curation.

Rosalie Bordett: Investigation, Writing - review and editing

Chi Zhang: Formal analysis, Writing - review and editing

Sukrut Hemant Karandikar: Investigation, Writing - review and editing

Rishikesh Pandey: Conceptualization, Investigation, Resources, Supervision, Methodology, Funding acquisition, Writing – original draft.

Ishan Barman: Conceptualization, Resources, Investigation, Supervision, Methodology, Formal analysis, Funding acquisition, Writing - original draft, Data curation.

Declaration of interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

References

  1. Ahmad A, Dubey V, Singh VR, Tinguely J, Øie CI, Wolfson DL, Mehta DS, So PT, Ahluwalia BS, 2018. Lab on a Chip. 18, 3025–3036. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Ayyappan V, Chang A, Zhang C, Paidi SK, Bordett R, Liang T, Barman I, Pandey R, 2020. ACS sensors. 5, 3281–3289. [DOI] [PubMed] [Google Scholar]
  3. Basu S, Kolouri S, Rohde GK, 2014. Proceedings of the National Academy of Sciences. 111, 3448–3453. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Bergholt MS, Serio A, Albro MB, 2019. Frontiers in bioengineering and biotechnology. 7, 303. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Boustany NN, Boppart SA, Backman V, 2010. Annu. Rev. Biomed. Eng 12, 285–314. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Breiman L, 2001. Mach. Learning 45, 5–32. [Google Scholar]
  7. Chhetri RK, Phillips ZF, Troester MA, Oldenburg AL, 2012. PLoS One. 7, e49148. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Demšar J, Curk T, Erjavec A, Gorup Č, Hočevar T, Milutinovič M, Možina M, Polajnar M, Toplak M, Starič A, 2013. The Journal of Machine Learning Research. 14, 2349–2353. [Google Scholar]
  9. Di Z, Klop MJ, Rogkoti V, Le Dévédec SE, van de Water B, Verbeek FJ, Price LS, Meerman JH, 2014. PloS one. 9, e109688. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. El-Mashtoly SF, Yosef HK, Petersen D, Mavarani L, Maghnouj A, Hahn S, Kötting C, Gerwert K, 2015. Anal. Chem 87, 7297–7304. [DOI] [PubMed] [Google Scholar]
  11. Ember KJ, Hoeve MA, McAughtrie SL, Bergholt MS, Dwyer BJ, Stevens MM, Faulds K, Forbes SJ, Campbell CJ, 2017. NPJ Regenerative medicine. 2, 12. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Good Z, Sarno J, Jager A, Samusik N, Aghaeepour N, Simonds EF, White L, Lacayo NJ, Fantl WJ, Fazio G, 2018. Nat. Med 24, 474. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Hamada K, Fujita K, Smith NI, Kobayashi M, Inouye Y, Kawata S, 2008. J. Biomed. Opt 13, 044027. [DOI] [PubMed] [Google Scholar]
  14. Hosseini P, Abidi SZ, Du E, Papageorgiou DP, Choi Y, Park Y, Higgins JM, Kato GJ, Suresh S, Dao M, 2016. Proceedings of the National Academy of Sciences. 113, 9527–9532. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Jo Y, Cho H, Lee SY, Choi G, Kim G, Min H, Park Y, 2018. IEEE Journal of Selected Topics in Quantum Electronics. 25, 1–14. [Google Scholar]
  16. Jolliffe, I, 2002. Principal component analysis. Wiley Online Library. [Google Scholar]
  17. Karandikar SH, Zhang C, Meiyappan A, Barman I, Finck C, Srivastava PK, Pandey R, 2019. Anal. Chem 91, 3405–3411. [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Kasprowicz R, Suman R, O’Toole P, 2017. Int. J. Biochem. Cell Biol 84, 89–95. [DOI] [PubMed] [Google Scholar]
  19. Kim K, Kim KS, Park H, Ye JC, Park Y, 2013. Optics express. 21, 32269–32278. [DOI] [PubMed] [Google Scholar]
  20. Kim K, Park WS, Na S, Kim S, Kim T, Do Heo W, Park Y, 2017. Biomedical optics express. 8, 5688–5697. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Kim K, Yoon H, Diez-Silva M, Dao M, Dasari RR, Park Y, 2013. J. Biomed. Opt 19, 011005. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Kumamoto Y, Mochizuki K, Hashimoto K, Harada Y, Tanaka H, Fujita K, 2019. The Journal of Physical Chemistry B. 123, 2654–2661. [DOI] [PubMed] [Google Scholar]
  23. Lam VK, Nguyen T, Phan T, Chung B, Nehmetallah G, Raub CB, 2019. Cytometry. 95, 757–768. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Lee S, Park H, Kim K, Sohn Y, Jang S, Park Y, 2017. Scientific reports. 7, 1–11. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Lieber CA, Mahadevan-Jansen A, 2011. Appl. Spectrosc 57, 1363–1367. [DOI] [PubMed] [Google Scholar]
  26. Managò S, Valente C, Mirabelli P, Circolo D, Basile F, Corda D, De Luca AC, 2016. Scientific reports. 6, 24821. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Movasaghi Z, Rehman S, Rehman IU, 2007. Applied Spectroscopy Reviews. 42, 493–541. [Google Scholar]
  28. Nordlund J, Bäcklin CL, Zachariadis V, Cavelier L, Dahlberg J, Öfverholm I, Barbany G, Nordgren A, Övernäs E, Abrahamsson J, 2015. Clinical epigenetics. 7, 11. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Okada M, Smith NI, Palonpon AF, Endo H, Kawata S, Sodeoka M, Fujita K, 2012. Proceedings of the National Academy of Sciences. 109, 28–32. [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Paidi SK, Pandey R, Barman I, 2020. Encyclopedia of Analytical Chemistry, 1–21.
  31. Paidi SK, Pandey R, Barman I, 2020. Molecular and Laser Spectroscopy, 623–652.
  32. Paidi SK, Shah V, Raj P, Glunde K, Pandey R, Barman I, 2021. Biosensors and Bioelectronics, 112863. [DOI] [PMC free article] [PubMed]
  33. Pandey R, Zhou R, Bordett R, Hunter C, Glunde K, Barman I, Valdez T, Finck C, 2019. Journal of biophotonics. 12, e201800291. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Park Y, Depeursinge C, Popescu G, 2018. Nature Photonics. 12, 578–589. [Google Scholar]
  35. Pavillon N, Hobro AJ, Akira S, Smith NI, 2018. Proceedings of the National Academy of Sciences. 115, E2676–E2685. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Phillip JM, Han K, Chen W, Wirtz D, Wu P, 2021. Nature Protocols. [DOI] [PMC free article] [PubMed]
  37. Popescu G, Park Y, Lue N, Best-Popescu C, Deflores L, Dasari RR, Feld MS, Badizadegan K, 2008. American Journal of Physiology-Cell Physiology. 295, C538–C544. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. Ross ME, Zhou X, Song G, Shurtleff SA, Girtman K, Williams WK, Liu H, Mahfouz R, Raimondi SC, Lenny N, 2003. Blood. 102, 2951–2959. [DOI] [PubMed] [Google Scholar]
  39. Schie IW, Rüger J, Mondol AS, Ramoji A, Neugebauer U, Krafft C, Popp J, 2018. Anal. Chem 90, 2023–2030. [DOI] [PubMed] [Google Scholar]
  40. Shu X, Sansre S, Jin D, Zeng X, Tong K, Pandey R, Zhou R, 2020. arXiv preprint arXiv:2012.08518.
  41. Winnard PT Jr, Zhang C, Vesuna F, Kang JW, Garry J, Dasari RR, Barman I, Raman V, 2017. Oncotarget. 8, 20266. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Woo JS, Alberti MO, Tirado CA, 2014. Experimental hematology & oncology. 3, 1–14. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Wu P, Phillip JM, Khatau SB, Chen W, Stirman J, Rosseel S, Tschudi K, Van Patten J, Wong M, Gupta S, 2015. Scientific reports. 5, 1–10. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1
NIHMS1714887-supplement-1.pdf (78.5KB, pdf)

RESOURCES