一种有效的图像处理方法对贮存红细胞闪烁复杂性的评估

刘 丹,周晓宇,艾 渤,官 科,李 楠

(1.中国信息通信研究院,北京100191; 2.北京交通大学轨道交通控制与安全国家重点实验室,北京100044; 3.中兴通讯股份有限公司,北京100191)



一种有效的图像处理方法对贮存红细胞闪烁复杂性的评估

刘 丹1,周晓宇1,艾 渤2,官 科2,李 楠3

(1.中国信息通信研究院,北京100191; 2.北京交通大学轨道交通控制与安全国家重点实验室,北京100044; 3.中兴通讯股份有限公司,北京100191)

摘 要:体外红细胞老化研究具有极高的临床价值.临床应用上的不足,是由于缺少从微观帧上获取大量闪烁数据的有效方法.研究了一种新的图像处理方法,从不同的微观帧上提取红细胞灰度强度变化情况,使得细胞分裂能够在一张图像上显示.尺度不变特征转换被用来移除摄动的细胞.通过多尺度样本熵对来自25个年轻成年人(每人200个细胞)的红细胞膜闪烁的复杂性进行分析,同时,也测量三磷酸腺苷浓度.结果显示,提出的图像处理方法可靠且能够有效提取红细胞闪烁强度,红细胞老化过程能够通过膜闪烁的复杂性分析检测到,该过程比三磷酸腺苷测量要敏感得多.

关键词:闪烁;分裂;老化;红细胞;多尺度样本熵

Blood component transfusion drastically promoted the clinical studies on the preservation of the RBC.Currently,the RBCs are usually conservation with Alsever’s solution at 4℃[1].Under this in-vitro circumstance,RBC is exposed to the environment completely different from that of the in-vivo condition;depriving of the blood circulation,loss of plasma and the other blood cells and paralysis of the normal signaling pathways and the purgation mechanisms for metabolic wastes or senescent cells.It can at least account for the reason that the storage period of RBC is only about 35-42 days compared with its in-vivo lifespan as 120±4 days[2].

There were massive clinical reports on the adverse consequences of transfusing older storage-age RBCs[3],which necessitated the new evaluation for the quality of the stored RBCs.Conventional biochemical methods[4]included the reduction of ATP level,2,3-diphosphoglycerate(2,3-DPG),p H and glycolysis rate,the accumulation of lactic acid and the increase in intracellular Ca2+.Many of these indicators associated with the cell damage.Since the change of the mechanical and the rheological properties of RBC was initiated on the cytomembrane and could induce a cascading degradation of the stored RBC quality[5],clinical practice demanded for a new method based on the biophysical metrics to evaluate the in-vitro function of RBC prior to the vesiculation of the RBC membrane and the release of the intracellular matters.

In this background,Costa et al.[6]and Szekely et al.[7]applied the multi-scale sample entropy(MSE)[8]to evaluate the flickering[9]complexity of the RBCs and found that old RBCs had a lower complexity than the young ones.Flickering is a typical oscillation on the membranes of the RBCs,which was reported decades ago[9].This activity is regulated by cytomembranes and cytoskeletons which are crucial to the cell function and viability[10].Thus,although the cited studies have not accurately measured the changes during the different phases of the storage,flickering is a promising predictor for assessing the RBC’s aging process.

Membrane flickering is the undulation of the pixel-based grayscale intensities in the microscopic video.Therefore,intensity values from millions of pixels in thousands of frames needed to be acquired and computed.So,the key point in evaluating the complexity of the flickering was developing a reliable and efficient method to extract the flickering data from the time-series images.By now,no efficient procedures have been addressed to process the massive data,which hindered the application of the method.As such,the available studies were based on the results from a small amount of the cells.For example,Costa et al.[6]processed 13 cells and Szelely et al.[7]computed the intensity variation of the cellular centremost pixels to generate the entropic results.On the other hand,to accurately present the flickering values,the cells were preferred to be fixed in the field of view during imaging.Any affine transformation(such as translation,rotation,shear deformation scaling,etc.)should be rigorously prevented.Otherwise,the flickering data would be polluted by the errors from the spatial drift.To track the cells,the model based method[11],the spatiotemporal analysis[12],the mean-shift algorithm[13]and the Scale-Invariant Feature Transform(SIFT)[14]have ever been applied.

In this study,we developed a template based method to identify the membrane flickering from the frame sequence.The RBCs on only one image were segmented automatically.SIFT was used to screened out the perturbed RBCs.The positions of the segmented and the unperturbed cells were transferred to the other frames.Accordingly,flickering extraction was conducted on these identified cells.It could significantly save the efforts in delineating the cells in the image sequence.Based on the work,the MSE analysis was performed.Statistical results showed that the applied methodology could detect the aging process of the stored RBC.ATP measurement confirmed the MSE results,which in return,demonstrated the validity of the proposed image processing method.It also showed that the flickering complexity was sensitive enough for assessing the aging process of the stored RBC.

1 Materials and Methods

1.1 Sample preparation

Twenty-five healthy male university senior students(non-smoker,20.32±1.29 year-old,and free of cardiovascular diseases)were recruited in the experiments as the blood donors.The venous blood was drawn in the anticoagulant tubes(Vacutainer,Becton Dickinson,Franklin Lakes,NJ,USA)and transferred to the centrifuge tubes with 0.9% NaCl.The white blood cells and the platelets were removed by centrifuging at 2 000 rpm/min for 5 mins.After removal of the supernatants,RBCs were re-suspended in the Alsever’s solution at 4℃.

1.2 Image acquisition

The images were taken in a darkroom with a vibrationproof table(63-541,TMC,Peabody,MA,USA).The images of the RBC were taken by a high resolution CCD camera(ORCA-ER2,Hamamatsu Photonics K.K.,Hamamatsu,Japan)via a microscope(BX-61,Olympus,Tokyo,Japan)coupled with a live-cell station.The sampling frequency was 70 Hz,two times of the maximum flickering frequency[15].The images took the resolution as 512×672 pixels.Grayscale level from 0 to 4095 was used to discrete the flickering intensity(212levels).The selection of the parameters was in agreement with the analysis of RBC membrane fluctuation using the digital holographic microscopy[16].The imaging system is shown in Fig.1.

The cells were placed in the live-cell station 10 mins before the image acquisition so that the RBCs could precipitate.Totally 6000 consecutive frames(86 s)have been taken.The initial and the rear images were removed to avoid the possible disturbance by the switch on/off of the instruments and the movement of the operators.

1.3 Extracting the flickering data

The principle of this part of work was to locate the cells in the individual video frame and to pick up the grayscale intensities per pixel on the frame sequence.

Manual segmentation with thousands of frames was very time-consuming.Recently,automated tools for cell segmentation have been reported[17].We developed and optimized our image processing procedures based on the features of the specific images.Fig.2(a)shows a typical frame for the RBC grayscale images.The RBCs had spherical-like(disk)contours and clear boundaries in sharp contrast to the background.The cells were well sedated before the image acquisition but the affine transformation might occur and the drifted cells needed to be eliminated.

1.3.1 Segmentation for the cell template

Segmentation was performed only on one raw frame shown in Fig.2(a).Sobel operator[18]was used to detect the cell edge shown in Fig.2(b).The morphological operations as dilations and erosions(kernel array:3×3)were conducted shown in Fig.2(c).Subsequently,the histogram for the connected area was calculated to determine the thresholds shown in Fig.2(d)for Otsu two-value segmentation[19].In this step,the cell clusters or the abnormal cells could also be removed.The segmented template for the positions of the RBCs is shown in Fig.2(e).Fig.2(f)displays the delineated cells in the raw image.

1.3.2 Dismissing the perturbed cells

We applied the SIFT tracking to remove the perturbed RBCs.This method was robust to the variation of image scaling,rotation,illumination,noise and even the viewpoint[14].It was uniquely conducted on the delineated cells in the constructed template.

By the analysis,the image was transformed to the scale space by convolving with multi-scale Gaussian functions

L(x,y,kσ)=G(x,y,kσ)*I(x,y)(1)where L(x,y,kσ)is the image in the scale space; G(x,y,kσ)is the Gaussian kernel functions at scale kσ,k=1,2,…,n;I(x,y)is the original grayscale image.

We calculated the difference by

A pixel-based comparison was made with its 26 neighbor pixels(8 in the same scale,9 in the two adjacent scales distinctively)to locate the maxima and minima as the SIFT key-points.Consequently,the selected key-points were assigned with descriptors which were used for calculating the minimum Euclidean distance.For example,one cell on two randomly selected images is shown and its key-points were matched in Fig.3,the lines indicate the correspondence of the key-points on two different images.Tab.1 lists the co-ordinations of the matched key-points in the two images.

Inconsistent co-ordinations in the two images indicated that the cell was drifted or deformed.We set a tolerance for recognizing the change of the x-y co-ordinations,for example,5%,below which,we believed that the variation was insignificant and no aforementioned affine transformation occurred for the studied RBC.

The perturbed cells were eliminated from the cell template and the remained cells were deemed to be well sedated during imaging.

Tab.1 Co-ordinations of the key-points by SIFT tracking on different images

1.3.3 Computing the flickering intensity

The co-ordinations from the segmented and the unperturbed cells on the constructed template were delivered to the other frames in the video sequence.The flickering intensity was subsequently performed

where int(x,y,t)represents the grayscale intensity of the pixel(x,y)on the tthimage;(x,y)was from the co-ordinations transferred from the segmented and the unperturbed cells;fint(x,y,t)is the tthflickering value on pixel(x,y).

To note,the calculation was only conducted on the pixels belonging to the selected cells on the template.This simplification involved a reduction of the calculation cost normally up to 80%.

1.4 MSE evaluation

Assuming that the acquired n-point flickering value of one pixel was saved in a data group x(n)= {x(1),x(2),…x(n)},we then re-arranged the data group into m-dimension subsets

Then,the similarity of two different subsets was calculated by

For each i,result of D[xm(i),xm(j)]was compared with r(a predefined tolerance,which was frequently defined as 0.2 time of the standard deviation of the flickering intensity[20]).

When the calculation has been completed for i,the normalized value was computed by

Then the same work has been repeated for the subsets with dimension m + 1 to obtain Bm+1(r).

The sample entropy was computed by

The MSE method was developed for multiscale variability over a range of scales[21].So,the n-point series{x(1),x(2),…x(n-1),x(n)} could be re-sampled(coarse-graining)with the scale s=s1

where floor denotes round toward the nearest inte-ger.

Then,a new time series {y(1),y(2),…y[floor(n/s)]},was constructed for evaluating the sample entropy.

1.5 ATP concentration measurement

ATP concentration was measured using ATP Assay Kit(S0026,Beyotime Institute of Biotechnology,Shanghai,China)for the RBCs immediately after the blood separation and on the 1st,3rd,5th,7th,10th,15thday after separation.

1.6 Statistical analysis

We averaged the entropy results across 200 cells per subject for six scales.Paired t-test was conducted to compare the complexity differences between the samples immediately after the blood separation and with different periods of conservation.The threshold with a significance level of 0.05 was applied.Time changes of ATP contents was evaluated using the paired t-test as well.

2 Results

2.1 Efficiency and the validity of the method

The image processing time for 6000 frames(constructing the template,eliminating the perturbed cells and extracting the flickering intensities for 200 RBCs)cost 267 s.The MSE calculation lasted 1 447 s for 200 RBCs.A personal computer(CPU:Intel Xeon E5;Memory:8 GB DDR3; Hard Disk:250 GB)was used in the calculation.

Fig.4 displays the pixel-based flickering intensity on a well sedated cell and a rotated cell identified by the SIFT based cell screening method.The rim-shape flickering pattern in Fig.4(a)was similar to the previous studies[6-7,9,15]and confirmed the effectiveness of our proposed image acquisition/processing procedures.

The rotated cell did not change its contour but the calculated flickering intensity was so irregular as to mislead the MSE results.It showed the importance for the inclusion of the cell screening method.

2.2 MSE results

Statistical results demonstrated the decrease of the MSE with the storage time.We listed the paired t-test results for each scale in Tab.2.

Tab.2 Statistical results of the MSE for different storage time

S1-S6represents the sample entropy on six scales and D0-D15represents the different time of storage(from the time immediately after the blood separation to 15thday after blood separation).We observed that the significant decrease in entropy occurred firstly in the lower scales and then in the high scales.The significant difference was found on the 3rdday of conservation.From the 5thday of conservation,the significant decrease in all the scales was found.

2.3 ATP measurement

In contrast,the significant decrease in ATP content(p＜0.05)was only observed from the 5thday after blood separation.

3 Discussions

The applied method could acquire and compute the pixel-based flickering MSE values from 200 cells in around 30 m using a regular personal computer.The time cost was comparable to a regular blood test.In addition,we used the SIFT based cell tracking to remove the perturbed cells.As such,the flickering complexity results were much reliable.In general,the method was suitable for the further high-throughput clinical trials in terms of its efficiency and effectiveness.

MSE values decreased with the elapse of the time.It was consistent with results of Costa[6]et al.on the new formed and the aged RBC.Complexity of a biological system reflected its ability to adapt to and to function in a changing environment[22].Entropy is the measure for the complexity.In particular,lower entropy of the flickering implied a predictable and regular fluctuation pattern on the membrane which associated with the membrane elasticity.As known,membrane elasticity dominated the squeezing ability of the RBCs inside the capillary vessels.Therefore,the decrease of the RBCs membrane elasticity(with the storage time)would raise the resistance of the blood capillary during the passage of the RBCs.As a consequence,RBCs’ability for carrying normal metabolism was weakened and toxicity in some organs could probably occur.In conclusion,using the entropy of membrane flickering was an appropriate metric to reflect the cell aging process.

Significant reduction in ATP was observed after 5 days of storage.This reduction correlated with the decrease of the MSE values but with a time delay.ATP transports chemical energy for cell metabolism.The energy could be used to synthesize the cellular material,to maintain the membranes,to fuel the movement and to activate the transportation[23].Metabolic trauma or stress(as being exposed to the in-vitro blood bank or storage lesion)may increase the consumption for ATP and reduce its regeneration.An important drop in cellular ATP will lead to cellular necrosis[24].In general,the reduction in ATP and in MSE of membrane flickering has similar meaning.Since the membrane fluctuations of the RBCs were regulated by multiple factors as bending modulus,membrane tension,and cytosolic viscosity,it was reasonable to see no punctual synchronization between the reduction in flickering MSE and in the ATP level[25].In general,it was beneficial to have different approaches to evaluate the RBC aging process.In particular,flickering MSE was sensitive compared with the ATP content.

We firstly observed the significant entropic difference for the lower scales,which corresponded to the undulations at higher frequencies.The effect was reasonable:the pixels on the membrane were actually dependent and they were scaffold by the cytoskeleton.The high-frequency flickering was mainly spontaneously originated by a few adjacent pixels(or a sub-domain)whilst the low-frequency variation usually involved the synchronization of much more pixels,which tended to neutralize the changes on the local region of the cell.In our study,the inclusion of the high-scale values was beneficial because they could provide the additional information to validate the results and the coarsegraining process(high scale values)could filter out the potential outliers[20,26].

Notably,it was necessary to point out that the physique and healthy condition including the fatigue state,some haematological and cardiovascular diseases might yield abnormal results[27].Donors with close ages and similar backgrounds wereselected in our experiments.It was important to achieve a consistent result.

4 Conclusions

1)The aging process of the stored RBC can be effectively evaluated by MSE analysis with the membrane flickering.

2)The efficient image acquisition,data processing and evaluation procedures for massive flickering data have been developed,which enabled its practical application.ATP measurement has also been performed to confirm the MSE results.

3)The comparison showed that the MSE of membrane flickering was sensitive for reflecting the cell aging process and could be a promising method in clinical practice.

References:

[1]ALSEVER J B,AINSLIE R B.A new method for the preparation of dilute blood plasma and the operation of a complete transfusion service[J].New York State Journal of Medicine,1941,41:126-135.

[2]ANTONELOU M H,KRIEBARDIS A G,PAPASSIDERI IS.Aging and death signaling in mature red cells:from basic science to transfusion practice[J].Blood Transfusion,2010,8(S3):39-47.

[3]KOCH C G,LI L,SESSLER D I,et al.Duration of red-cell storage and complications after cardiac surgery[J].New England Journal of Medicine,2008,358(12):1229-1239.

[4]D’ALESSANDRO A,LIUMBRUNO G,GRAZZINI G,et al.Red blood cell storage:the story so far[J].Blood Transfusion,2010,8(2):82-88.

[5]CLUITMANS J C A,CHOKKALINGAM V,JANSSEN A M,et al.Alterations in red blood cell deformability during storage:a microfluidic approach[J].Bio Med Research International,2014:764268.

[6]COSTA M,GHIRAN I,PENG C K,et al.Complex dynamics of human red blood cell flickering:alterations with in vivo aging[J].Physical Review E,2008,78(2):020901.

[7]SZEKELY D,YAU T W,KUCHEL P W.Human erythrocyte flickering:temperature,ATP concentration,water transport,and cell aging,plus a computer simulation[J].European Biophysics Journal,2008,38(7):923-939.

[8]SARKER A,BARAT P.Multiscale entropy analysis:a new method to detect determinism in a time series[J].ar Xiv Preprint Physics,2006:0604040.

[9]BLOWERS R,CLARKSON E M,MAIZELS M.Flicker phenomenon in human RBCs[J].The Journal of Physiology,1951,113(2/3):228-239.

[10]SPARROW R L.Time to revisit red blood cell additive solutions and storage conditions:a role for“omics”analyses[J].Blood Transfusion,2012,10(S2):7-11.

[11]KACHOUIE N N,FIEGUTH P,RAMUNAS J,et al.A model-based hematopoietic stem cell tracker in image analysis and recognition[DB/OL].Springer Berlin Heidelberg,2005:861-868.

[12]BEAUMONT L M A,WAKEFIELD J,NOBLE J A.Spatiotemporal Bayesian cell population tracking and analysis with lineage construction[J].IEEE International Symposium on Biomedical Imaging,2008,5:340-343.

[13]DEBEIR O,VAN H P,KISS R,et al.Tracking of migrating cells under phase-contrast video microscopy with combined mean-shift processes[J].IEEE Transactions on Medical Imaging,2005,24(6):697-711.

[14]LOWE D G.Object recognition from local scale-invariant features[DB/OL].Proceedings of the International Conference on Computer Vision 2,1999:1150-1157.

[15]BROCHARD F,LENNON J F.Frequency spectrum of the flicker phenomenon in erythrocytes[J].Journal de Physique,1975,36(11):1035-1047.

[16]RAPPAZ B,BARBULl A,HOFFMANN A D,et al.Spatial analysis of erythrocyte membrane fluctuations by digital holographic microscopy[J].Blood Cells,Molecules,and Diseases,2009,42(3):228-232.

[17]MEIJERING E.Cell segmentation:50 years down the road[J].Signal Processing Magazine,2012,29(5):140 -145.

[18]HADWIGER M,KNISS J M,REZK-SALAMA C,et al.Real-time volume graphics[M].Ak Peters,2006:112-114.

[19]LIAO P S,CHEN T S,CHUNG P C.A fast algorithm for multilevel thresholding[J].Journal of Information Science and Engineering,2001,17(5):713-727.

[20]COSTA M,GOLDBERGER A L,PENG C K.Multiscale entropy analysis of biological signals[J].Physical Review E,2005,71(2):021906.

[21]LI J,ZHANG J,GE W,et al.Multi-scale methodology for complex systems[J].Chemical Engineering Science,2004,59(8):1687-1700.

[22]PENG C K,COSTA M,GOLDBERGER A L.Adaptive data analysis of complex fluctuations in physiologic time series[J].Advances in Adaptive Data Analysis,2009,1(1):61-70.

[23]MIKIROVA N,RIORDAN H D,KIRBY R K,et al.Monitoring of ATP levels in red blood cells and t cells of healthy and ill subjects and the effects of age on mitochondrial potential[J].Journal of Orthomolecular Medicine,2005,20(1):50-58.

[24]DE KORTE D,HAVERKORT W A,VAN GENNIP A H,et al,Nucleotide profiles of normal human blood cells determined by high-performance liquid chromatography[J].Analytical Biochemistry,1985,147(1):197-209.

[25]EVANS J,GRATZER W,MOHANDAS N,et al.Fluctuations of the red blood cell membrane:relation to mechanical properties and lack of ATP dependence[J].Biophysical Journal,2008,94(10):4134-4144.

[26]COSTA M,GOLDBERGER A L,PENG C K.Multiscale entropy analysis of complex physiologic time series[J].Physical Review Letters,2002,89(6):1021-1024.

[27]CICCO G,PIRRELLI A.Red blood cell(RBC)deformability,RBC agreeability and tissue oxygenation in hypertension[J].Clinical Hemorheology and Microcirculation,1999,21:169-177.

Evaluating the flickering complexity of the stored red blood cells using an efficient image processing method

LIU Dan1,ZHOU Xiaoyu1,AI Bo2,GUAN Ke2,LI Nan3
(1.China Academy of Information and Communications Technology,Beijing 100191,China; 2.State Key Laboratory of Rail Traffic Control and Safety,Beijing Jiaotong University,Beijing 100044,China; 3.Zhongxing Telecommunication Equipment Corporation,Beijing 100191,China)

Abstract:Aging of the in-vitro red blood cell(RBC)is a topic with great clinical interests.The gap for its clinical application was the lack of the efficient and effective method to obtain the massive flickering data from the microscopic frames.A new image processing method was developed to extract the grayscale intensity variation of the RBCs from different frames.By the method,the cell segmentation could be performed on one image.Scale-invariant feature transformation was used to remove the perturbed cells.The flickering complexity of the RBCs from 25 young adults(200 cells for each)was analyzed by multi-scale sample entropy.Meanwhile,adenosine triphosphate(ATP)concentration was measured.The results showed that the proposed image processing method was reliable and efficient to extract the flickering intensity of the RBCs.The aging process of the stored RBCs could be detected using the complexity analysis of the membrane flickering,which was much sensitive than ATP measurement.

Key words:flickering;segmentation;aging;red blood cells;multi-scale sample entropy

作者简介:刘丹(1985—),女,河北石家庄人,工程师,博士生.研究方向为生物电磁学.email:liudan@caict.ac.cn.

基金项目:国家自然科学基金资助项目(61371187)

收稿日期:2015-11-23

DOI:10.11860/j.issn.1673-0291.2016.01.015

文章编号:1673-0291(2016)01-0092-09

中图分类号:Q64

文献标志码:A