Optical Imaging for Diagnosis of Rheumatoid Arthritis

Automatic Versus Human Evaluation

Pouyan Mohajerani

, Reinhard Meier

, Ernst J. Rummeny

and Vasilis Ntziachristos

Institute for Biological and Medical Imaging, Technische Universität München and Helmholtz Zentrum München,

Ingolstädter Landstrasse 1, Neuherberg 85764, Germany

Department of Radiology, Klinikum Rechts der Isar, Technische Universität München, München, Germany

Keywords: Optical Imaging, Fluorescence, Rheumatoid Arthritis (RA), Inflammation, Indocynine Green (ICG), Planar

Illumination, Near-Infrared Dyes, Spatiotemporal Analysis, Principal Component Analysis (PCA).

Abstract: Successful detection of rheumatoid arthritis (RA) at the early stages of development can significantly

enhance the chances of effective therapy. The early onset of RA is often marked with inflammation of the

synovial lining of the joint, a condition known as synovitis. Effective imaging of synovitis is therefore of

critical importance. While dynamic, contrast-enhanced magnetic resonance imaging (MRI) is capable of

effective imaging of synovitis, it is a costly modality. As an alternative, inexpensive approach, optical

imaging post injection of the near-infrared fluorescent dye indocynine green (ICG) has been recently

proposed for imaging RA. Evaluation of the obtained optical images is performed via examination by

trained human readers. However, optical imaging has yet to achieve the diagnostic accuracy of MRI. In this

paper we present a method for automatic evaluation of the fluorescence images and compare its

performance with the human-based evaluation. Our method relies on our previous work on spatiotemporal

analysis of image sequence with principal component analysis (PCA) to seek synovitis signal components

with the help of a segmentation method. The results for a group of 600 joints, obtained from 20 patients,

suggest improved diagnostic performance using the automatic approach in comparison to human-based

evaluation.

1 INTRODUCTION

Imaging can play a critical role in developing

effectively and timely therapeutic approaches for

treating rheumatoid arthritis (RA) by the way of

early detection of synovitis (Emery and Quinn,

2003, Ostergaard et al., 2005). Synovitis is the

condition of the inflammation of the synovial lining

surrounding the joint and marks the onset of RA.

Conventionally, X-ray computed tomography (CT)

has been employed to image bone and joint damage

resulting from joint inflammation (Backhaus et al.,

1999). In this sense CT is often applicable in the

later stages of RA development. Other anatomical

modalities, particularly magnetic resonance imaging

(MRI) in conjunction with MR contrast agents and

ultrasound have been employed for early detection

of RA (Emery et al., 2007). Nevertheless, such

methods are often limited by factors such as operator

dependency for ultrasound (Delle Sedie et al., 2008)

and high costs for the MRI, (Emery et al., 2007)).

As an alternative, cost-effective approach,

optical imaging (OI) has been proposed for imaging

RA (Chen et al., 2005, Hielscher et al., 2004, Fischer

et al., 2010, Mohajerani et al., 2013, Mohajerani et

al., 2014, Meier et al., 2012, Gompels et al., 2010).

Several planar and tomographic approaches have

been proposed, relying on the physiological changes

in joint tissue as a source of optical contrast

(Hielscher et al., 2011, Klose et al., 1999).

The application of fluorescence has been also

recently proposed for imaging RA in both planar

(Meier et al., 2012, Werner et al., 2012) and

tomographic (Mohajerani et al., 2014) modes. These

approaches use the organic, near-infrared

fluorescence dye indocynine green (ICG) to create

optical contrast in affected tissue. Specifically, the

intravenously injected ICG tends to accumulate in

inflamed synovial tissue, a feature which enables

differentiation of the inflamed joints from healthy

joints using measured fluorescence signals. In

particular, planar imaging operating in epi-

illumination mode offers the ability to image all

hand joints in both hands simultaneously (Meier et

al., 2012, Meier et al., 2014, Werner et al., 2012).

Mohajerani P., Meier R., Rummeny E. and Ntziachristos V..

Optical Imaging for Diagnosis of Rheumatoid Arthritis - Automatic Versus Human Evaluation.

DOI: 10.5220/0005217200360043

In Proceedings of the International Conference on Bioimaging (BIOIMAGING-2015), pages 36-43

ISBN: 978-989-758-072-7

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

The acquired fluorescence images are examined

by a trained human reader, who assigns semi-

quantitative inflammation scores to different joints

upon examining the entire acquired image sequence.

An example an acquired fluorescence image in

comparison to the corresponding MRI scan is

presented in Figure 1.

Planar, ICG-enhanced imaging of RA is a

promising approach which has been applied for both

diagnostic (Thomas Dziekan et al., 2011, Meier et

al., 2012) as well as therapy monitoring purposes

(Meier et al., 2014). While offering a cost-effective

and rapid imaging alternative, the diagnostic

accuracy of the method is compromised, in

particular in comparison with contrast-enhanced

MRI, which serves as the gold standard in evaluation

studies (Meier et al., 2012).

One reason for the relatively (compared to MRI)

low diagnostic accuracy of OI, is the presence of

strong signal interference (Meier et al., 2012).

Specifically, the fluorescence signal emanating from

the underlying synovitis is strongly coupled to other

signals originating from the dorsal veins as well as

other tissues. This interference occurs in both spatial

and temporal dimensions and complicates the

inference process as performed by a human reader.

We have previously reported a spatiotemporal

analysis approach for decoupling the signal

components in the acquired image sequence using

principal component analysis (PCA) (Mohajerani et

al., 2013).

In this paper, we present automatic detection of

the synovitis, as an alternative to the human-based

detection. We further evaluate and compare the

automatic detection method with the human-based

detection for a cohort consisting of 20 patients.

2 IMAGING METHODOLOGY

Fluorescence imaging was performed with a planar,

near-infrared fluorescence imaging system (Xiralite

X4, Mivenion GmbH, Berlin, Germany. This system

enables real-time image acquisition at the

fluorescence wavelength of ICG (around 830 nm)

simultaneously from both of the hands after epi-

illumination excitation applied to the dorsal hand

sides. The patients received a bolus injection pf ICG

at a dosage of 0.1 mg per kg body weight. A total of

360 images were obtained at the fluorescence

wavelength of ICG, with a frame rate of 1 fps.

Figure 1: Optical imaging of rheumatoid arthritis shown in

the left panel for the left hand of a 42 year old male

patient. This patient exhibited severe arthritis in the 3rd

metacarpophalangeal (MCP) joint. The right panel depicts

the corresponding transversal slice of the T1-weighted fat-

saturated contrast-enhanced MRI image in the MCP

region. The higher accumulation of the contrast agents in

the MCP 3 joint have resulted in higher signal intensities

in both MRI and optical images.

3 SPATIOTEMPORAL ANALYSIS

Non-specific fluorescence signal interfering with the

target fluorescence emanating from synovitis

complicated the diagnosis. The interference takes

place both in the intra-frame domain (spatial

interference) as well as in the inter-frame domain

(temporal interference). Decoupling such signal

components in both spatial and temporal domains

might therefore help with more accurate diagnosis.

We have recently presented spatiotemporal

analysis for decomposing the signal components in

the fluorescence image sequences (Mohajerani et al.,

2013). This method makes use of the principal

component analysis (PCA) (Jolliffe, 2002), as an

orthogonal de-correlating transformation, to covert

the original sequence into a group of sequences,

each bearing distinct spatiotemporal components.

Herein we briefly review this approach.

Specifically, consider a set of fluorescence

images I

for p = 1 … P, where each image has a

size of M

× M

pixel size (P was equal to 360).

Prior to the PCA analysis, two levels of localization

were performed on the raw image sequence I

. The

first step limits the processing to a specific region

within each image, achieved via a region of interest

defined accordingly for each of the joints, as shown

in Figure 2. The second localization step confines he

PCA processing to windowed subsequence of the

OpticalImagingforDiagnosisofRheumatoidArthritis-AutomaticVersusHumanEvaluation

Figure 2: The proposed methodology of automatic detection of rheumatoid arthritis (a) partitioning the fluorescence image

into 7 regions for the carpus, MCP region and each of the fingers. (b) PCA-based processing and segmentation-based

detection within each ROI (here, the carpus ROI). The results of the PCA processing are to then fed to the segmentation

approach. The metric values assigned by the segmentation method to different signal components are then classified to

achieve diagnostic information regarding each joint.

images within each ROI. These two localization

steps were performed to improve the performance of

PCA in decoupling image components, as the signal

dynamics change significantly across time and

space. For each of the 7 ROIs shown, a 2-D

subsequence of images J

is defined from the

original sequence. Next, K successive images were

taken from the subsequence J

. This subsequence is

defined as H

, i = 1 to K. The PCA was then applied

to this image sequence by first vectorising and

stacking the images H

, to achieve a K × M

matrix X

. Next, singular value decomposition

(SVD) of the covariance matrix of X was obtained.

and the resulting unitary matrix of its eigenvectors

was applied as the PCA transformation matrix. The

3 components with the largest singular values were

then retained and the rest of modes were discarded.

These 3 retained components were then mapped to

the blue, red and green channels of an output color

sequence, according to their descending singular

values. Further details of the PCA processing can be

found in (Mohajerani et al., 2013). A summary of

the steps involved can be further seen in figure 2.

The results of the spatiotemporal analysis for a

specific joint of patient suffering severe

inflammation in the 3

MCP joint are presented in

figure 3. As observed, the PCA-based method has

successfully separated the synovitis signal from the

background and vein signals into distinct PCA

channels.

4 AUTOMATIC DETECTION OF

RHEUMATOID ARTHRITIS

As previously noted, the synovitis signal is often

coupled in time and space to interfering signals

emanating from background tissue or dorsal veins.

However, the synovitis signal is likely to appear as a

distinct component in one of the PCA channels.

Here, we propose automatic detection of

synovitis by searching through the PCA channels for

a signal component attributable to synovitis. To this

end, we use a segmentation approach previously

proposed in (Mohajerani et al., 2013).

Specifically, for a specific joint, first an elliptical

ROI is defined surrounding this joint. This ROI is

denoted by the binary image R. The segmentation

approach then applies a threshold to each PCA

image. The thresholding results in a binary image J.

Within the image J, the connected component with

the largest overlap with R is found and denoted by

K. A metric is defined then to quantify the likelihood

of the signal in the region K to be due to synovitis.

Specifically, S(K,R) is defined as

BIOIMAGING2015-InternationalConferenceonBioimaging

()

(,) 3

max ( , ),

1max ,(,),1 () ,

SKR

dKR r

JKR EK

=×



−−





(1)

where J(K, R) and d(K, R) denote respectively the

Jaccard (Michael Levandowsky and Winter, 1971)

and Hausdorff (Huttenlocher et al., 1993) distances

between K and R and E(K) is the energy of the

image I within the label in K (Mohajerani et al.,

2013).

Figure 3: An example of the PCA-based spatiotemporal

processing applied to the MCP-region of a patient with

severe synovitis in the MCP 3 joint of the right hand. (a)

Shows the measured raw fluorescence image, where the

synovitis signal is coupled with background and dorsal

vein signals. (b) The decomposed image obtained using

the PCA-based approach, where the synovitis signal is

mapped to the blue channel and is clearly distinct from the

vein signals mapped to the green and red channels.

The metric S(K,R) has a value between 0 and 3,

where a higher value denotes a higher likelihood that

the contours of K delineate the synovitis signal in the

corresponding PCA channel.

The automatic detection of the synovitis then

operates as follows. All PCA images for all the blue,

red and green channels are then processed with the

segmentation method presented and the

corresponding metric values are found. The

component K with the highest value of S(K,R) across

all channels in then designated as the synovitis

signal. A summary of this approach is presented in

figure 2. Figure 4 shows the results of the automatic

detection method for a specific case of a patient with

moderate synovitis in the left carpus. In this specific

case, the component with the highest metric value

appeared in the red channels, as shown by the black

contour in figure 4(b).

Figure 4: Case study of the automatic detection method

proposed herein for the left carpus joint of a 49 year old

female patient with moderate synovitis. (a) The blue and

red PCA components for the carpus ROI, where the black

contours denotes the detected synovitis signal in the red

channel. The vein signal was mapped to the blue channel

and there was no conspicuous signal in the green channel.

(b) The corresponding raw fluorescence image. The scores

assigned by the 4 human readers to this joint consisted of

0, 0, 0 and 1.

5 CLINICAL EXAMINATION

AND COHORT INFORMATION

The development presented and performed in this

paper have been conducted in the context of a recent

study carried out at the Klinikum rechts der Isar,

Munich, Germany, which aimed at of evaluating the

diagnostic performance of ICG-aided imaging of RA

(Meier et al., 2012).

The automatic detection method proposed herein

was applied to fluorescence image sequences

Table 1: Distribution of inflammation severity among the

600 hand joints of the 20 patients recruited in this study.

Healthy Mild Moderate Severe

Carpus 16 16 5 3

MCP 93 90 16 1

PIP/DIP 326 28 5 1

OpticalImagingforDiagnosisofRheumatoidArthritis-AutomaticVersusHumanEvaluation

obtained from a group of 20 patients (14 females, 6

males, aged 41±16). The patients were examined

and imaged with contrast-enhanced MRI using a 3T

MR machine (Verio, Siemens Erlangen, Germany)

and a protocol described in (Meier et al., 2012). MR-

based synovitis scores of 0 to 3 (healthy to severe)

were assigned to each of the 600 joints.

Three radiologists scored the degree of

inflammation in a total of 30 joints of both hands

using the MR scans. Synovitis scores on a 4-point-

ordinate scale (0: no inflammation, 1: mild, 2:

moderate, 3: severe) were assigned to each joint

according to the semi-quantitative assessment

system suggested by the OMERACT MRI group

(Ostergaard et al., 2003). The MR scores constitute

the true diagnostic information, as explained in the

next section.

Similarly, the fluorescence images were scores

by the 3 radiologists, with a repeated 4

scoring

performed after 4 weeks, as explained in (Meier et

al., 2012). As such, 4 scores between 0 and 3, are

obtained for each of the 30 hand joints for each

patient. These scores are then used to evaluate

human-based evaluation of the optical images, in

comparison to the proposed, automatic method. The

distributions of synovitis severity within different

joint groups (interphalangeal, metacarpophalangeal

and carpal) are shown in Table 1, according to the

examination results of the MR scans.

6 EVALUATION METHODOLOGY

The localization metric devised in Section 4 yields a

value between 0 and 100 (with 100 designating

highly likelihood of being an inflammation signal) to

each signal component. A threshold can be applied

to this localization metric toward making a decision

about synovitis severity of a given joint. The results

can be demonstrated using the so-called receiver

operating characteristic (ROC) curves, denoting as

sensitivity vs specificity.

ROC curves denote the classification

performance of a binary classifier (Zou et al., 2007).

It should be noted that the x = y on the ROC plane

corresponds to random classification. Therefore, any

curve above this line is desirable. The optimal

performance corresponds to the upper-left corner

(sensitivity = specificity = 100%). One way to

compare different ROC curves is to compare the

area under curve (AUC). The AUC is a measure of a

classifier’s quality (Fawcett, 2006). The optimal

classifier has an AUC of 1.

To achieve a binary classifier, we consider two

modes of classification. In classification 1, the

threshold is applied to the segmentation metric to

make a decision between healthy and affected (mild,

moderate or severe synovitis) joints. In classification

II, a decision was made between joints with “no or

mild synovitis” and joints with “moderate or severe

synovitis”.

Three diagnosis methods are examined and

evaluated in this work:

• Method A: Human evaluation of raw images

• Method B: Automatic evaluation of raw images

• Method C: Automatic evaluation of PCA images

Method A is the conventional method used

currently in the clinic (Meier et al., 2012). Method C

constitutes the proposed method. We have already

shown using a cohort of 15 patients that the method

C outperforms method B (Mohajerani et al., 2013).

Figure 5: Receiver operator characteristic (ROC) curves

for classification I: healthy joints versus affected joints

(mild, moderate or severe synovitis). The ROC curves are

shown using three methods: segmentation of the raw

images (blue curve), segmentation of the PCA components

(red curve, the proposed method) and the human-based

evaluation (black curve). The green, dotted vertical line

denotes a detection specificity of 60%. For this value, the

proposed approach outperforms the human evaluation by

achieving a sensitivity of more than 70% (human

evaluation had a sensitivity of around 55%). For higher

values of specificity (more than 70%), all methods showed

poor sensitivity of less than 50%.

In this paper, we examine the performance of the

proposed method (method C), in comparison with

the human-based read (method A), for the first time.

BIOIMAGING2015-InternationalConferenceonBioimaging

The results are presented in the next section. For

method A, the synovitis score was taken as the

medium of the 4 scores obtained by the 4 readers (as

explained in Section 2). For methods B and C, the

synovitis score was the metric S, shown in Eq. 1.

7 DIAGNOSTIC PERFORMANCE

RESUTLS

Figure 5 shows the ROC curves for classification I,

as defined in the previous section. The proposed

method (method C, red curve) showed slightly better

performance than the other two methods for the

specificity value of 60% (marked with filled circles).

Over all, both automatic and human-based detection

had an equal performance in terms of the AUC

values, as shown in Table 2.

Table 2: Area under curve (AUC) values for the three

detection methods for classification I (healthy vs. affected)

and for classification II (healthy or mild synovitis vs.

moderate or severe synovitis).

Method A Method B Method C

Classification I 0.65 0.57 0.65

Classification II 0.77 0.73 0.82

Figure 6: ROC curves for classification II: (healthy joint

or mild synovitis) versus (moderate or severe synovitis),

presented similar to the results presented for classification

I in Figure 5. The green, dotted vertical line denotes a

detection specificity of 60%. For this value, the proposed

approach outperforms the human evaluation by achieving

a sensitivity of more than 95% (human evaluation had a

sensitivity of around 75%).

Figure 6 presents similar results for classification

II. In this case, all methods perform better than the

corresponding curves in classification I. This is

justified as the signal difference between moderate

synovitis and healthy or mild synovitis is generally

larger than between healthy and mild synovitis. The

proposed method C in this case has a markedly

better performance than human reading. This

improved performance can be seen examining the

sensitivity values for a medium specificity value of

60% (as marked with filled circles on figure 6) or

examining the AUC values, presented in Table 2.

8 DISCUSSION AND

CONCLUSION

In this paper we presented automatic detection and

characterization of synovitis in human hand joints

using fluorescence images obtained in epi-

illumination geometry post intravenous injection of

ICG. The proposed method scores the principal

components obtained from spatiotemporal analysis

of the raw image sequences. The scores are then

used to classify the synovitis in a binary fashion.

Two classification modes were examined:

classification I (differentiating between healthy and

affected joints) and classification II (differentiating

between healthy or mild synovitis vs. moderate or

severe synovitis). The automatic evaluation of

fluorescence images was compared with the current

methodology of human-based evaluation of images

for both classification scenarios using ROC curves,

as shown in Figures 5 and 6. The comparison of the

proposed method with human-based reading was

performed for a cohort consisting of 20 patients.

For classification I, as can be seen in Figure 5,

all three methods A, B, and C have relatively low

sensitivity. The reason is that in classification I we

differentiate between healthy and affected joints.

Around 22% (according to Table 2) of all joints have

mild arthritis. However, mild joints exhibit only very

low signal contrast relative to the healthy joints. This

issue lowers the sensitivity of all methods. However,

the proposed method has better sensitivity than the

human detection for medium specificity levels. For

classification II, however, the proposed method

outperformed human detection in both sensitivity

and AUC terms.

It should be noted that for both classification I

and II, the methods A and B have higher sensitivity

than the proposed method, for very high specificity

values (> ~80%), as seen in Figures 5 and 6.

However, these higher sensitivity values are

OpticalImagingforDiagnosisofRheumatoidArthritis-AutomaticVersusHumanEvaluation

generally lower than 50%, which translate to a high

count of missed positives for the respective

threshold values.

The results generally show improved or

comparable diagnostic performance achieved using

the proposed, automatic method in comparison to

the human-based evaluation. It is foreseeable that

better and more intelligent classification methods

making use of all signal properties (and not just the

extracted segmentation metric S) could lead to

definitively better performance than human reading.

Such improved detection can not only lead to better

detection but can also improve the therapy

monitoring utility of optical imaging by reducing

operator dependency. Such improved classification

is being currently researched. Furthermore, semi-

quantitative scoring of synovitis using optical

images as well as further development of the

proposed method in conjunction with larger cohorts

are subjects of ongoing work.

REFERENCES

Backhaus, M., Kamradt, T., Sandrock, D., Loreck, D.,

Fritz, J., Wolf, K. J., Raber, H., Hamm, B., Burmester,

G. R. & Bollow, M. 1999. Arthritis Of The Finger

Joints: A Comprehensive Approach Comparing

Conventional Radiography, Scintigraphy, Ultrasound,

And Contrast-Enhanced Magnetic Resonance

Imaging. Arthritis Rheum, 42, 1232-45.

Chen, W. T., Mahmood, U., Weissleder, R. & Tung, C. H.

2005. Arthritis Imaging Using A Near-Infrared

Fluorescence Folate-Targeted Probe. Arthritis

Research & Therapy, 7, R310-R317.

Delle Sedie, A., Riente, L. & Bombardieri, S. 2008. Limits

And Perspectives Of Ultrasound In The Diagnosis

And Management Of Rheumatic Diseases. Modern

Rheumatology, 18, 125-131.

Emery, P. & Quinn, M. A. 2003. Window Of Opportunity

In Early Rheumatoid Arthritis: Possibility Of Altering

The Disease Process With Early Intervention. Clinical

And Experimental Rheumatology, 21, S154-S157.

Emery, P., Wakefield, R. J., O'connor, P. J., Conaghan, P.

G., Mcgonagle, D., Hensor, E. M. A., Gibbon, W. W.

& Brown, C. 2007. Finger Tendon Disease In

Untreated Early Rheumatoid Arthritis: A Comparison

Of Ultrasound And Magnetic Resonance Imaging.

Arthritis & Rheumatism-Arthritis Care & Research,

57, 1158-1164.

Fawcett, T. 2006. An Introduction To Roc Analysis.

Pattern Recognition Letters, 27, 861–874.

Fischer, T., Ebert, B., Voigt, J., Macdonald, R., Schneider,

U., Thomas, A., Hamm, B. & Hermann, K. G. 2010.

Detection Of Rheumatoid Arthritis Using Non-

Specific Contrast Enhanced Fluorescence Imaging.

Acad Radiol, 17, 375-81.

Gompels, L. L., Lim, N. H., Vincent, T. & Paleolog, E. M.

2010. In Vivo Optical Imaging In Arthritis--An

Enlightening Future? Rheumatology (Oxford), 49,

1436-46.

Hielscher, A. H., Kim, H. K., Montejo, L. D., Blaschke,

S., Netz, U. J., Zwaka, P. A., Illing, G., Muller, G. A.

& Beuthan, J. 2011. Frequency-Domain Optical

Tomographic Imaging Of Arthritic Finger Joints. Ieee

Trans Med Imaging, 30, 1725-36.

Hielscher, A. H., Klose, A. D., Scheel, A. K., Moa-

Anderson, B., Backhaus, M., Netz, U. & Beuthan, J.

2004. Sagittal Laser Optical Tomography For Imaging

Of Rheumatoid Finger Joints. Physics In Medicine

And Biology, 49, 1147-1163.

Huttenlocher, D. P., Klanderman, G. A. & Rucklidge, W.

J. 1993. Comparing Images Using The Hausdorff

Distance. Ieee Transactions On Pattern Analysis And

Machine Intelligence, 15, 850-863.

Jolliffe, I. T. 2002. Principal Component Analysis,

Springer.

Klose, A. D., Hielscher, A. H., Hanson, K. M. & Beuthan,

J. 1999. Two- And Three-Dimensional Optical

Tomography Of Finger Joints For Diagnostics Of

Rheumatoid Arthritis. Proceedings Of Spie, 3566,

151-160.

Meier, R., Thuermel, K., Moog, P., Noel, P., Ahari, C.,

Sievert, M., Dorn, F., Waldt, S., Schaeffeler, C.,

Werner, S., Golovko, D., Haller, B., Ganter, C.,

Weckbach, S., Woertler, K. & Rummeny, E. J. 2012.

Detection Of Arthritis In The Hands Of Patients With

Rheumatological Disorders: Diagnostic Performance

Of Optical Imaging In Comparison To Magnetic

Resonance Imaging. Arthritis & Rheumatism, 64,

2489-98.

Meier, R., Thuermel, K., Noël, P. B., Moog, P., Sievert,

M., Ahari, C., Nasirudin, R. A., Golovko, D., Haller,

B., Ganter, C., Wildgruber, M., Schaeffeler, C., Waldt,

S. & Rummeny, E. J. 2014. Synovitis In Patients With

Early Inflammatory Arthritis Monitored With

Quantitative Analysis Of Dynamic Contrast-Enhanced

Optical Imaging And Mr Imaging. Radiology, 270,

176-185.

Michael Levandowsky & Winter, D. 1971. Distances

Between Sets. Nature, 234, 34-35.

Mohajerani, P., Koch, M., Thürmel, K., Haller, B.,

Rummeny, E. J., Ntziachristos, V. & Meier, R. 2014.

Fluorescence-Aided Tomographic Imaging Of

Synovitis In The Human Finger. Radiology.

Mohajerani, P., Meier, R., Noël, P. B., Rummeny, E. J. &

Ntziachristos, V. 2013. Spatiotemporal Analysis For

Indocyanine Green-Aided Imaging Of Rheumatoid

Arthritis In Hand Joints. Journal Of Biomedical

Optics, 18, 097004-097004.

Ostergaard, M., Ejbjerg, B. & Szkudlarek, M. 2005.

Imaging In Early Rheumatoid Arthritis: Roles Of

Magnetic Resonance Imaging, Ultrasonography,

Conventional Radiography And Computed

Tomography. Best Practice & Research In Clinical

Rheumatology, 19, 91-116.

BIOIMAGING2015-InternationalConferenceonBioimaging

Ostergaard, M., Peterfy, C., Conaghan, P., Mcqueen, F.,

Bird, P., Ejbjerg, B., Shnier, R., O'connor, P.,

Klarlund, M., Emery, P., Genant, H., Lassere, M. &

Edmonds, J. 2003. Omeract Rheumatoid Arthritis

Magnetic Resonance Imaging Studies. Core Set Of

Mri Acquisitions, Joint Pathology Definitions, And

The Omeract Ra-Mri Scoring System. Journal Of

Rheumatology, 30, 1385-1386.

Thomas Dziekan, Carmen Weissbach, Jan Voigt, Bernd

Ebert, Rainer Macdonald, Malte L. Bahner, Marianne

Mahler, Michael Schirner, Michael Berliner, Birgitt

Berliner, Jens Osel & Osel, I. 2011. Detection Of

Rheumatoid Arthritis By Evaluation Of Normalized

Variances Of Fluorescence Time Correlation

Functions. Journal Of Biomedical Optics 16, 076015

Werner, S. G., Langer, H. E., Ohrndorf, S., Bahner, M.,

Schott, P., Schwenke, C., Schirner, M., Bastian, H.,

Lind-Albrecht, G., Kurtz, B., Burmester, G. R. &

Backhaus, M. 2012. Inflammation Assessment In

Patients With Arthritis Using A Novel In Vivo

Fluorescence Optical Imaging Technology. Annals Of

The Rheumatic Diseases, 71, 504-510.

Zou, K. H., O'malley, A. J. & Mauri, L. 2007. Receiver-

Operating Characteristic Analysis For Evaluating

Diagnostic Tests And Predictive Models. Circulation,

115, 654-7.

OpticalImagingforDiagnosisofRheumatoidArthritis-AutomaticVersusHumanEvaluation