Distinct Fluorescence Optical Imaging Patient Clusters Emerge for Seropositive and Seronegative Rheumatoid Arthritis

Abstract

Objective

To investigate whether digital activity fluorescence optical imaging (FOI) patterns of inflammation can identify distinct rheumatoid arthritis (RA) phenotypes.

Methods

The hands of newly diagnosed patients with RA were evaluated by clinical examination, musculoskeletal ultrasound, and FOI. Inflammation on FOI was defined when capillary leakage and/or fluorophore perfusion was present. The FOI composite image was quantified into a digital disease activity (DACT) score, using novel computerized algorithms. Unsupervised clustering on FOI inflammatory patterns was used to identify subgroups of patients relative to anticyclic citrullinated peptides (ACPA) and/or rheumatoid factor (RF).

Results

Of 1326 examined hand joints in 39 patients with RA (72% female; 56% ever‐smokers; 54% RF positive and 69% ACPA positive), 400 (30%) showed inflammation by FOI, and 95% (37 of 39) of patients had DACT‐FOI scores greater than 1. Unsupervised analysis on FOI patterns revealed two patient clusters, cluster 1 (n = 29) and cluster 2 (n = 10). The proportion of seropositive patients was significantly higher in cluster 1 versus cluster 2 (90%, 26 of 29 vs. 30%, 3 of 10; P < 0.01), whereas C‐reactive‐protein levels (minimum‐maximum) were significantly higher in cluster 2 (20 mg/l [1‐102]) versus cluster 1 (2 mg/l [0‐119]; P = 0.01). A wider variety and proportion of inflamed joints emerged for patients with RA in cluster 2 versus cluster 1, in which inflammation was more concentrated around the wrists and the right metacarpophalangeal 2 (MCP2), bilateral MCP3, and, to a lesser degree, left MCP2 and proximal interphalangeal joint and tendon regions. Cluster 1 displayed lower mean (±SD) DACT scores compared with cluster 2 (3.6 ± 2.1 vs. 5.4 ± 2.1; P = 0.03).

Conclusion

FOI‐based digital quantification of hand joint inflammation revealed two distinct RA subpopulations with and without ACPA and RF related autoantibodies.

INTRODUCTION

Detection of abnormal rheumatoid arthritis (RA)–related autoantibodies, such as rheumatoid factor (RF) and anticitrullinated protein antibody (ACPA), together with advanced clinical diagnostic imaging and certain acute‐phase reactant tests, plays an important role toward earlier diagnosis, risk estimation, and testing of new therapeutic strategies in early RA subpopulations (1, 2). There is currently a lack of objective measures that can help differentiate among RA phenotypes, and the heterogeneity of the disorder can sometimes offer challenges in diagnosis and management in clinically difficult patients. Apart from established imaging techniques that uses x‐rays, magnetic resonance, nuclear medicine, and musculoskeletal ultrasound (MSUS), there is also a more recent fluorescence optical imaging (FOI) technology that is gaining recognition as an operator‐independent diagnostic tool for early undifferentiated and/or inflammatory arthritis and may serve useful in the management of RA (3, 4, 5).

FOI, or “Rheumascan” (Xiralite X4, Mivenion GmbH), is an emerging modality designed for the hands that use an intravenous fluorophore (indocyanine green [ICG]) to identify altered microcirculation in areas of abnormally high perfusion and/or capillary leakage (3, 4, 5, 6, 7, 8). The concordance between clinical and ultrasound arthritis has been reported in several articles (9, 10, 11, 12, 13); however, there are relatively limited FOI studies in patients with newly diagnosed RA. Our previous studies showed FOI sensitivity and specificity to range between 73% and 84% and 83% and 95% in detecting clinically manifest or silent synovitis, with positive and negative predictive values of 77% and 97%, respectively, in those with various rheumatic diseases (6). FOI may have higher sensitivity than MSUS for detecting altered microcirculatory changes of inflamed hand tissues in small joints (8) and may contribute to timely detection of subtle inflammatory changes in and around the joint capsule, tendons, and/or skin tissue of patients with rheumatic diseases (4, 5, 6, 8). FOI may also have the potential to monitor the effects of therapy using reliable semiquantitative FOI scoring methods (14, 15, 16, 17).

We have employed FOI in previous reports by analyzing the images by a computer‐generated algorithm of signal intensity (SI) patterns represented as an automated digital disease activity (DACT) score that has proved technically feasible and reproducible with a good‐to‐moderate agreement between MSUS and clinical measures of disease activity in patients with rheumatic diseases (6, 8, 18). Here, we used cluster analyses of FOI images to discern hand joint inflammatory patterns in patients newly diagnosed with RA.

PATIENTS AND METHODS

Clinical examination, ultrasound imaging, and radiography

Participants selected for the FOI study were attendants (suspected of RA development) at the early arthritis clinic of the Karolinska University Hospital in Stockholm, Sweden, during years 2013‐2016. Patients were evaluated by standard clinical examination (CE) and MSUS and offered an FOI examination when clinically justified by the referring physician (6, 8, 18). Study approval was obtained from the Karolinska Institute ethics board. The clinical, MSUS, and FOI examinations were usually carried out the same day. Only those fulfilling the RA 2010 American College of Rheumatology/European Alliance of Associations for Rheumatology (ACR/EULAR) classification criteria (1) that received a RA diagnosis according to the rheumatologist were included in this study.

Blinded by the clinical and FOI examinations, one of three examiners performed the MSUS (hands and feet, to include symptomatic joints) using instrument presets according to Rezaei et al (6, 8, 18) on the General Electric LOGIQ E9 ultrasound machine. Active synovitis was defined as synovial hypertrophy (grade ≥2) with Doppler signaling (grade ≥1), and only binary scores for positive and negative inflammation were used for this study analysis (6). The presence of tenosynovitis was also documented. Bone erosions were identified on radiographs by reporting radiologists.

FOI semiquantitative scoring (visual‐FOI)

The FOI examination was performed using standardized procedure (6, 8, 14). Hands were washed prior to examination to maintain uniform tissue temperature, ensuring artifact‐free images. An intravenous bolus (fluorophore ICG pulsion, 0.1 mg/kg body weight) was administered 10 seconds prior to procedure, and patients were asked to restrict finger movements under dark‐room conditions throughout the 6‐minute examination (Supplementary File 1).

Blinded by the clinical and MSUS results, the FOI images were evaluated using consensus scoring by two inspectors with FOI experience (specialist rheumatologist [EaK] and health care professional [YK]). Abnormally increased focal optical signal intensities or enhancements in areas of high perfusion and/or capillary leakage were scored as FOI positive. The composite image and entire FOI series were first evaluated in “rainbow” and then in “temperature” palette formats to help distinguish normal vasculature (narrow, tubular, and well‐marginated structural intensities) from inflammatory tissue enhancements. The 34 joint regions of interest (three wrist joints: radiocarpal, midcarpal, and distal radioulnar joint regions; and five metacarpophalangeal [MCP], one interphalangeal [IP], four proximal interphalangeal [PIP], and four distal interphalangeal [DIP] hand joints) were visually inspected for joint tissue inflammation and were graded according our modified semiquantitative evaluation published elsewhere (14) (Supplementary File 1). For this current study, only binary scores that marked the presence or absence of joint inflammation were analyzed.

DACT quantitative scoring

The DACT score is an FOI computer‐generated algorithm (XiraView software Vi.4.0) of the composite image (240 frames per second summary) that was automatically quantified for each patient.

Automatic extraction of the hands from the image background was performed by computer software (19). The algorithm separated the hands from the forearm in a standardized manner by using a multiple of the length of the middle finger (size corrected and calibrated). The DACT‐FOI formula was based on fluorescence intensity curve thresholds to discriminate intensity variations (high and low areas). The area of high SI was automatically calculated, and the values of each patient were divided by the 95th percentile of intensities in healthy individuals as the reference value (8,19).

DACT calculations used the above algorithm to determine the normal threshold that was automatically assigned a numerical score of DACT‐FOI less than 1 regarded as the normal digital activity SI range (8, 19).

Data acquisition

An independent research physician not involved in the scoring acquired the clinical joint counts, MSUS and FOI reports, and data charts from patient journals for analysis.

Statistical methods

IBM SPSS V.26.0 and GraphPad Prism 6.0 software were used for statistical analyses. Unsupervised ascending hierarchical clustering and principal component analysis were used to identify inflammatory FOI patterns of joint involvement. The robustness of the clustering was verified using k‐means, and an agreement between both methods was assessed using κ statistics. Baseline clinical and biological characteristics of patients were compared between the clusters using nonparametric testing.

RESULTS

A total of 1326 hand joints of 39 newly diagnosed patients with RA were studied. Baseline patient characteristics are given in Table 1. The mean (±SD) symptom duration was reported as 18.2 (±21.7) months. Of the 39 patients, 74% were seropositive (69% ACPA positive, whereas 54% were RF positive), of whom 19 (49%) had both RF and ACPA positivity (Supplementary Table 2). The mean (minimum‐maximum) results for the erythrocyte sedimentation rate (ESR) and C‐reactive protein (CRP) levels were 30.2 (3‐86) mm/hour and 22.9 (0‐119) mg/l, respectively, and 56% of patients (19 of 29 for seropositive RA and 6 of 10 for seronegative RA) were current or previous smokers (ever‐smokers).

Table 1.

Patient characteristics, laboratory results, and assessment methods

Baseline characteristics	All patients (N = 39)
Sex
Female, n (%)	28 (72)
Age at diagnosis, mean (±SD), y	56 (17.4)
Symptom duration, mo	18.2 (21.7)
ACPA status, n (%)
Positive	27 (69)
Negative	12 (31)
RF status, n (%)
Positive	21 (54)
Negative	18 (46)
ESR levels (mm/h) mean (minimum‐maximum)	30.2 (3‐86)
CRP levels (mg/l) mean (minimum‐maximum)	22.9 (0‐119)
Smoking status (ever smoking), n (%)	22 (56%)
Radiograph examination
Bone erosions	9 (23)
Osteoarthritis	12 (31)
Clinical examination (hands)
Patients swollen joints, n (%)	29 (74)
Joints swollen, mean (SD)	7.8 (8.1)
Ultrasound examination (hands)
Patients MSUS synovitis, n (%)	37 (95)
Joints MSUS synovitis, mean (SD)	9.7 (7.7)
FOI (hands)
Patients FOI inflamed joints, n (%)	37 (95)
Joints FOI inflamed, mean (SD)	10.3 (7.2)
DACT scoring (hands)
Patients DACT‐FOI positive, n (%)	37 (95)
DACT‐FOI score, mean (SD)	4.1 (2.1)

Abbreviations: ACPA, anticyclic citrullinated peptide antibody; CRP, C‐reactive protein; DACT, digital disease activity; ESR, erythrocyte sedimentation rate; FOI, fluorescence optical imaging; MSUS, musculoskeletal ultrasound; RF, rheumatoid factor.

Hand radiographs revealed 23% (9 of 39) of the total population had bone erosions, and five of nine patients with erosions were RF positive, whereas eight of nine were ACPA positive. The proportion of patients diagnosed with hand joint inflammation were 29 (74%) by CE, 37 (95%) by MSUS, and 37 (95%) by DACT‐FOI. The mean (±SD) number of joints inflamed were 7.8 (±8.1), 9.7 (±7.7), and 10.3 (±7.2) by clinical, MSUS, and FOI examination respectively, with a mean DACT score of 4.1 (±2.1).

Clustering of digitally active FOI patterns identify early RA phenotypes

Unsupervised hierarchical clustering of digitally active FOI patterns of inflammation identified two main patient clusters: cluster 1 (n = 29) and cluster 2 (n = 10).

A greater enrichment of digital FOI activity patterns emerged for patients toward the right of the dendrogram as compared with those on the left, and fewer subgroupings of seronegative patients were identified in cluster 1 (10% [3 of 29]) as compared with cluster 2 (70% [7 of 10]; P < 0.01). Significantly higher proportional differences emerged for subgroups of seropositive patients in cluster 1 compared with cluster 2 (90% [26 of 29] vs. 30% [3 of 10]; P < 0.01). More patients were ACPA positive in cluster 1 than in cluster 2 (83% [24 of 29] vs. 30% [3 of 10]; P = 0.004), whereas no differences between clusters for RF were observed (62% [18 of 29] vs. 30% [3 of 10]; P = 0.08). No significant differences for age, sex, smoking status, and ESR levels were noted between clusters (Figure 1).

Figure 1.

Dendrogram illustrates two distinct RA patients clusters generated by unsupervised hierarchical analysis of digitally active FOI joint inflammation patterns for subpopulations with and without ACPA and RF autoantibodies. The joint distribution graph on top left shows all 34 joint regions examined, illustrating percentage differences with probability values of joint inflammatory patterns. The table below displays proportional differences between clusters among the listed parameters for all 39 patients with RA. ACPA, anticyclic citrullinated peptide antibody; CRP, C‐reactive protein; DIP, distal interphalangeal; ESR, erythrocyte sedimentation rate; FOI, fluorescence optical imaging; MCP, metacarpal phalangeal; NS, not significant (P > 0.05); PIP, proximal interphalangeal; RA, rheumatoid arthritis; RF, rheumatoid factor.

Numerically more erosions were present for RA subpopulations in cluster 1 (seven patients) compared with cluster 2 (two patients; P > 0.05). However, distinct differences regarding elevated CRP levels (20 mg/l [1‐102] vs. 2 mg/l [2‐119]) and increased mean number (±SD) of inflamed joint distributions (11.6 ± 5.5 vs. 7.6 ± 5.3; P < 0.05) were seen in cluster 2 as compared with cluster 1.

Cluster differences were also verified using principal component analysis showing similar groupings between seropositive and seronegative RA subpopulations (Figure 2A).

Figure 2.

A, Principal component analysis of RA subpopulations grouped as cluster 1 (red) and cluster 2 (blue). B, This composite image is a summary of 240 combined images showing mean FOI DACT scores (DACT score <1) of a normal scan. C, Tabulations of FOI and DACT joint involvement patterns between clusters 1 and 2. D, Left is the summary image of DACT scores of a seropositive patient with RA in cluster 1 indicating inflammation (white arrows). The image on the right shows signal intensity enhancements (arrows) of multiple joints and inflamed tendons of a seronegative patient with RA in cluster 2. DACT, digital disease activity; FOI, fluorescence optical imaging; RA, rheumatoid arthritis.

Among the 400 inflamed joints, there were significant differences in inflammatory patterns between the two clusters (20% of 986 were inflamed in cluster 1 vs. 59% of 340 inflamed in cluster 2; P < 0.05) at the joint level. Higher proportions of digital activity pattern differences emerged for 64% of the 34 joint regions examined in cluster 2 as compared with cluster 1 (P < 0.05) (Figure 1 joint distribution graph). In cluster 1, joint inflammation was largely concentrated around wrists and the left MCP2, bilateral MCP3, and, to a lesser degree, PIP 2‐4 regions, whereas no significant inflammatory differences in the right MCP2, bilateral PIP5, or left MCP1 and DIP 2‐5 regions were seen between clusters.

DISCUSSION

Here we show how unsupervised hierarchical clustering analysis of DACT‐FOI patterns revealed two distinct clusters among RA phenotypes with and without ACPA and/or RF autoantibodies using the automated DACT‐FOI scoring method.

Like any novel modality used for arthritis detection, risks and benefits need to be thoroughly weighed prior to examination. Our experience with FOI suggests that it be used in an environment equipped to handle intravenous contrast procedures (6, 8, 18). This emerging modality offers no exposure to harmful radiation, it is relatively inexpensive, and images can be acquired rapidly (3, 4, 5, 6, 7, 8, 14). The localization of the signal intensities and enhancement patterns have shown to help differentiate the different types of arthritis (4, 5, 6, 14, 17).

About two thirds of all new RA cases are mainly characterized by the presence of ACPA and RF autoantibodies and have genetic environmental risk factors, cytokine profiles and histology differences, and more joint destruction when left untreated as compared with seronegative RA cases (20). Previous reports also showed that in the presence or absence of RA‐related autoantibodies (ACPA or RF), the utility of imaging, especially radiography, MSUS, and FOI accompanied by the automated DACT‐FOI analysis, reliably supports the rheumatologist's decision‐making (6, 8, 18, 19). Based on this RA subpopulation (Supplementary Figures 3 and 4), we showed the usefulness of FOI in identifying joint inflammation using ultrasound as a reference, to be 81% sensitive and 90% specific, with positive and negative predictive values of 96% and 61%, respectively (8). Notably, FOI was 68% sensitive in detecting silent synovitis defined as not inflamed on CE, but synovitis positive on ultrasound. Being among the first to add novel automated DACT scoring to our FOI assessments, we further showed reliable agreements with clinical and ultrasound evaluations, recommending its use in digitally quantifying hand joint inflammation in RA assessments (Supplementary File 2).

The generation of separate FOI patient clusters by unsupervised analysis showed significantly higher groupings of seropositive RA subpopulations with less systemic inflammation in cluster 1 versus cluster 2. The strength of unsupervised hierarchical clustering analysis is user independence and lack of selection bias; however, a limitation of performing cluster analysis on (many) joints instead of (few) patients exposes the risk of within‐patient nesting.

Although most of the patients were clinically diagnosed with early RA, there were nine that had a more erosive disease; seven of these nine were seropositive in cluster 1 compared with two of the nine with bone erosions identified in cluster 2, one of whom was diagnosed as having seronegative RA. Tendon involvement confirmed by ultrasound affected both seropositive (15 of 29) and seronegative (5 of 10) patients with RA, with tendon sheath inflammation (tenosynovitis) more prevalent in seropositive patients with RA (especially over the extensor‐carpi‐ulnaris‐wrist regions). Paratendon and/or joint capsule enhancement patterns mostly affected the second and third fingers of those in cluster 1 versus a wider variety of finger enhancement patterns over the paratendons for most of the patients in cluster 2.

The seronegative cluster 2 subpopulation reported a shorter symptom duration (8.8 ± 6.3 vs. 19.2 ± 3 months) with elevated CRP levels (80% vs. 48%; P < 0.05) as compared with cluster 1 subpopulation with seropositive RA. Also of note was that five of these seropositive patients (mean age <46 years) in cluster 1 self‐reported symptoms lasting up to 6 years without any signs of clinical and/or subclinical (ultrasound) arthritis prior to diagnosis, however, at the time of inclusion, FOI and ultrasound were able to confirm clinical signs of inflammation, and hence a seropositive RA diagnosis. Although an emphasis is placed on very early arthritis detection, it also needs to be highlighted that clinical arthritis by a rheumatologist is still the standard practice and over‐diagnosis by imaging methods alone be cautioned to prevent over‐treatment. Special precautionary measures were taken to avoid misinterpretations of imaging findings/misdiagnosis. An additional strength of our FOI study was that all patient reports were evaluated using consensus scoring by two experienced FOI examiners, one of whom being a senior rheumatologist.

A major limitation of our study was the small sample size, and a larger statistical power is warranted. It also needs to be considered that more inflamed joints in seronegative patients with RA could be confounded by reasons for meeting the 2010 ACR/EULAR RA classification criteria, in which greater than 10 arthritic joints were noted for patients with seronegative RA at diagnosis, which could have largely affected our study results. However, unsupervised clustering of clinical and ultrasound scores showed no distinguishable cluster differences. The identification of RA phenotypes using unsupervised clustering analysis of digital activity FOI patterns have, to our knowledge, not been shown before; however, it will require larger FOI data sets to offer more conclusive results.

Awareness of pitfalls when interpreting FOI images and quantifying pathological changes is important (5). With a substantial intrareader (κ = 0.73) and interreader (κ = 0.73) agreement (separate study), our semiquantitative FOI method of visual image inspections and interpretation proved reliable for arthritis scoring (14). The advantages of evaluating images in real time remain uncontested. Pattern recognition of individual images has limitations; however, a composition of 240 image frames summed up as a DACT score was an additional strength of our study. FOI together with the novel DACT scoring method shows good potential to complement a rheumatologist's clinical diagnosis (especially in clinically difficult cases in patients without known RA autoantibodies), as well as to help assess disease activity to objectively monitor the effects of therapy. A disconnection between improvement in disease activity and subsequent improvement in long‐term outcomes in RA without autoantibodies suggests that the underlying pathogenesis of seronegative and seropositive RA may be different (20).

Although unsupervised hierarchical clustering analysis of DACT on FOI revealed two distinct RA patient clusters for those with and without ACPA and RF autoantibodies, caution has to be taken in interpreting these data because of the relatively small sample size. Significantly higher groupings of seropositive patients with RA with lesser systemic inflammatory patterns concentrated more around the joints and tendons of wrists and the second and third finger regions emerged for cluster 1 versus cluster 2. The higher DACT scores, increased CRP concentrations, and a wider variety of inflammatory patterns spread diffusely across the wrist and finger regions emerged for most patients in cluster 2 compared with cluster 1, suggesting that FOI identifies patterns of joint involvement that are different for seropositive and seronegative RA.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be submitted for publication. Kisten had limited access, while Arnaud had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design

Kisten, Arnaud, af Klint, Rezaei.

Acquisition of data

Kisten, Levitsky, Györi, af Klint, Rezaei.

Analysis and interpretation of data

Kisten, Arnaud, af Klint, Rezaei.

Supporting information

Disclosure Form:

Appendix S1: Supporting Information