Cloud Hosting

This paper provides a proof-of-concept to use RT-DC for detection of MDS. As RT-DC captures the morphology of cells, the information content is similar to morphology analyses of BM smears which is currently the gold standard for MDS diagnosis. In addition, the mechanical readout of RT-DC is a promising feature as earlier studies showed alterations of the actin cytoskeleton in association with MDS8,9,10.

Current MDS diagnosis routines are under reconsideration due to reproducibility issues, high labor intensiveness, and requirement of expert staff2,4,19. These issues could be addressed utilizing a combination of imaging flow cytometry (IFC) for high-throughput acquisition and machine learning for automated data analysis20,21. In IFC, fluorescence images are captured which allows labelling of different cell types and intracellular structures. However it was already shown that using deep learning, brightfield images are sufficient for example to predict lineages of differentiation or distinguish cell types in blood22,23. Hence, the label-free approach of RT-DC could be advantageous as the staining process can be omitted.

In the present work we employ RT-DC for the first time for detection of MDS. From each captured cell, seven features are computed in real-time, which were then used to train a random forest model, reaching an accuracy of 82.9% for the classification of healthy and MDS samples. As RT-DC performs image analysis in real-time, the MDS classification result could be provided immediately during the measurement. Both, the label-free aspect of RT-DC and the real-time analysis could allow to shorten the time needed for diagnosis.

By employing a model interpretation technique, we found that the width of the distribution of cell sizes is one of the most important criteria used by the random forest classification model. While employing only a single feature for classification lowers the accuracy (78%), it may be more suitable for observation in clinical practice. Interestingly, our finding is in accordance with the WHO guidelines which suggest a consideration of cell sizes during morphology evaluation. Our measurements show consistently that a subpopulation of cells in the size range (25 mu {text{m}}^{2}le Ale 45 mu {text{m}}^{2}) is underrepresented in MDS samples (see Fig.1D and Supplementary Fig. S3). This effect could be explained by the reduced number of B lymphocyte precursor cells in MDS24, which are CD34+ and could be present in the sample after CD34 based sorting25. Moreover, the histogram of cell sizes in Fig.1D shows a narrow peak at 50m2 for MDS, while the healthy counterpart presents a wider distribution. Hence, especially the width of the distribution plays a role, rather than the mean or median which is similar for both samples. However, since only 41 samples have been employed to train and validate the random forest model, the extrapolation of this study on the highly heterogeneous MDS population is limited, as the model could be overfitted to this small dataset. Moreover, random forest models do not perform well in extrapolation tasks. Hence, a larger prospective clinical study is required to reach more decisive conclusions.

Our work considered seven features obtained using RT-DC which can be summarized into three groups: features describing cell size (A, Lx, Ly), mechanical properties (, D, I), and porosity (). However, updated versions of the RT-DC technology are capable to save the brightfield image and compute transparency features in real-time which was shown to allow for discrimination between different blood cell types26. Moreover, images can be evaluated by a deep neural net which allows to employ fine grained details of the image for an accurate classification22,23. Future research should incorporate those new modalities to improve label-free detection of MDS using RT-DC.

MDS is caused by accumulation of genetic mutations which can be identified by whole genome sequencing. While costs for whole genome sequencing reduced from a hundred million to a thousand dollars during the last 20years, currently only targeted sequencing plays a role in clinical practice27. Here, only chosen genes that are frequently affected in MDS are checked, which is problematic, due to the large genomic heterogeneity present in various types of MDS28,29. Therefore, the standard diagnosis relies on an assessment of cell morphology as an indirect readout of genetic properties. Morphological alterations are accompanied by changes in the F-actin distribution and structural changes of the cytoskeleton8,9,10. RT-DC allows to measure mechanical properties of cells that are determined by the cytoskeleton5,30,31. It was already shown that diseases like malaria, leukemia, or spherocytosis lead to measurable differences in mechanical properties26,32. To link mechanical and genetic changes, we measured HSCs from MDS patients using RT-DC and performed molecular analysis in parallel. Figure2B indicates that larger numbers of genetic mutations correspond to lower median deformation. Therefore, RT-DC could provide an additional indirect readout of acquired mutations that has low cost per measurement, low measurement time, and offers real-time analysis results. However, despite the high correlation, we would regard this finding as hypothesis-generating due to the small sample size (n=10). Additionally, we could neither identify an association of mutation type and deformation, nor a significant mechanical difference between the low and the high-risk group (data not shown), but rather the biological features of the blast cells, such as number of mutations, correlated with the mechanical properties. The importance of Dmedian resulting from the random forest model is low (see Fig.1B). This suggests that Dmedian is similar for healthy and MDS samples. Hence, the approach of correlating Dmedian to infer the number of mutations is only valid for samples for which MDS had already been diagnosed.

HSCs only make up approximately 1% of the cells in the bone marrow33,34. To focus our study on this small subpopulation, we used MACS for CD34 enrichment of HSCs prior to the measurement. However, as the cells produced by mutated HSCs are presumably morphologically different from the healthy counterpart, a future endeavor should assess unsorted bone marrow in RT-DC using a similar approach as shown in the present work. Moreover, the efficiency of CD34 isolation is low, which results in small total numbers of cells for the measurement. As a result, our measurements could not fully employ the available throughput-capacity of RT-DC. Samples shown in this manuscript were subjected to cryopreservation and thawing which could potentially alter the cell morphology and MDS prediction outcome. A follow up project should therefore ideally use fresh BM.

Taken together, our study shows that RT-DC has the potential to expand the current status quo of MDS diagnostics. Both, morphological and mechanical readout from RT-DC are promising parameters for identification of MDS. Whether this method can be complementary to the standard diagnostic procedures in the borderline cases or serve as a rapid reliable test in the initial diagnostics remains to be demonstrated in the prospective clinical studies.

See the article here:
Machine learning assisted real-time deformability cytometry of CD34+ cells allows to identify patients with myelodysplastic syndromes | Scientific...