To optimise the diagnostic process of rheumatic diseases affecting the hands using fluorescence optical imaging (FOI)

Abstract

Background

Accurate and rapid diagnosis of rheumatic diseases is essential for further treatment decision. Different rheumatic diseases present characteristic patterns (image features) in fluorescence optical imaging (FOI). We developed an atlas of FOI image features and tested its ability to differentiate various rheumatic diseases.

Methods

FOI images from patients with rheumatoid arthritis (RA), psoriatic arthritis (PsA), connective tissue diseases (CTD) and osteoarthritis (OA) were analysed by two readers blinded for diagnosis and calibrated against each other, using the prima vista mode (PVM) and an automated 5-phase model. Twenty-six different reoccurring typical signal enhancement patterns (features) indicating inflamed joints, nail or skin were defined and all FOI images were scored accordingly. The feature frequency in each patient cohort and phase (PVM, 5-phase) was counted. Contingency tables were created with categorical variable counts and diagnosis using common formulae.

Findings

Four hundred thirty-eight patients with RA (n=117), PsA (n=110), CTD (n=121) and OA (n=90) were included. Once the data had been categorised, a two-step diagnostic pathway was developed: in the first step, OA was best distinguished from the other diseases with high specificity by five patterns (specificity >0.9, diagnostic OR between 2.34 and 8.24). In a second step, the remaining autoimmune diseases were differentiated from each other by a certain number of features (five for RA, 12 for PsA and four for CTD).

Interpretation

This was the first study to show that feature analysis in FOI helps to differentiate typical rheumatic diseases from each other, potentially simplifying and speeding up the diagnostic process. Therefore, FOI could be considered an additional component of a wider range of imaging techniques used in rheumatology.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Accurate and rapid diagnosis of rheumatic diseases affecting the hands is essential for further treatment decision. In addition to MRI and musculoskeletal ultrasound, fluorescence optical imaging (FOI) is another sensitive method for the detection of inflammation of the hands. Different rheumatic diseases such as osteoarthritis (OA), rheumatoid arthritis, psoriatic arthritis and connective tissue diseases present characteristic patterns (image features) in FOI.

WHAT THIS STUDY ADDS

• Different reoccurring typical signal enhancement patterns (features) indicating inflamed joints, nail or skin were defined in an atlas of FOI image features and tested for their ability to differentiate various rheumatic diseases. In a two-step diagnostic pathway, OA was best distinguished from the other diseases, which were differentiated from each other in a second step.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• FOI helps to differentiate typical rheumatic diseases from each other, which could both speed up and improve the diagnostic process in the future. This could greatly improve patient care, particularly in rural areas where there are fewer rheumatologists, leading to earlier and more accurate diagnosis and earlier initiation of treatment.

Introduction

Accurate and rapid diagnosis of rheumatic diseases affecting the hands is essential for further treatment decision. For example, early treatment of rheumatoid arthritis (RA) has been shown to slow disease progression and positively influence its course.¹ Similarly, the early diagnosis and treatment initiation of psoriatic arthritis (PsA) is important to avoid long-term damage. At the same time, delayed diagnosis is associated with poorer physical function.^{2 3} This also applies to other rheumatic diseases affecting the hands such as connective tissue diseases (CTD).⁴ Current imaging techniques in rheumatology, in particular MRI and musculoskeletal ultrasound (MSUS) in grayscale and power Doppler mode, have long been established and are of great importance in the (early) diagnosis of autoimmune inflammatory joint diseases, such as RA and PsA. In contrast to conventional X-ray, MRI and MSUS are well suited for early diagnosis because they can detect synovitis, tenosynovitis and erosions at an early stage when they are still absent from X-ray;⁵ and they have a good inter-reader reliability. However, both techniques have their barriers, as they require a high level of expertise and, in the case of MRI, high costs, potential contraindications and limited availability.

In addition to MRI and MSUS, fluorescence optical imaging (FOI) is another sensitive method for the detection of inflammation of the hands.⁶ It visualises (early) inflammatory changes in various rheumatic diseases affecting the hands, for example, RA,^{7 8} PsA,⁹ CTD,^{10 11} osteoarthritis (OA)^{12 13} and others.¹⁴ Using indocyanine green (ICG) as a fluorescent optical dye, the software device visualises the microcirculatory changes caused by inflammation. Currently, the literature on FOI focuses mainly on the detection of joint inflammation (FOI Activity Score, FOIAS) and (only a few) on inflammatory skin changes.¹¹ Two recently published studies focus on feature reading, which has mostly been used for the diagnosis of CTD¹⁵ and PsA^{16 17} including, for example, the occurrence of Raynaud’s syndrome and enthesitis. However, there are also other structures that become visible on FOI examination and may be important in the differential diagnostic process of rheumatic diseases affecting the hands. Recently, 18 different structures have been presented as features on FOI.¹⁸

Therefore, the aim of this study is to differentiate typical rheumatic diseases affecting the hands (RA, PsA, CTD and OA) from each other by using predefined features in FOI. In a first step, we identified features to distinguish OA from the autoimmune inflammatory diseases (RA, PsA and CTD), and in a second step, we aimed to differentiate them from each other. With the presented analysis, we intend to simplify the early diagnostic process for rheumatic diseases affecting the hands by the identification of specific FOI criteria for each disease.

Methods

All FOI images were obtained as part of different observational studies at the Department of Rheumatology and Clinical Immunology, Charité—Universitätsmedizin Berlin, Germany. All studies have been approved by the German (mainly Berlin) ethical committees and patients had to give written consent to be included.

Study population

We analysed data from patients with rheumatic diseases affecting the hands, including RA, PsA, CTD and OA selected from different cohort studies (128/13 EK, 127/13 EK, EA1/045/10, EA1/193/10, EA1/269/13). Inclusion criterion was a confirmed diagnosis of either RA, PsA, CTD (mainly systemic sclerosis (SSc) and systemic lupus erythematosus (SLE)) and OA in adults by a rheumatologist based on the anamnesis, the clinical examination and laboratory parameters. Kidney function was within normal limits and hyperthyroidism was excluded before inclusion. Other exclusion criteria were hypersensitivity to fluorescent dyes, especially ICG, allergy to iodine, injured hands, pregnant or breastfeeding women, or significant uncontrolled or severe concomitant diseases.

Fluorescence optical imaging (FOI)

FOI was performed with the Xiralite system (Xiralite GmbH, Berlin, Germany) using a standardised procedure. FOI uses ICG as a fluorescent dye and LEDs for optical excitation (740 nm). After intravenous application of ICG (a bolus of 0.1 mg/kg), the dye shows distinct spatial temporal patterns or accumulates in the inflamed areas of both hands. This is visualised by near-infrared imaging and image processing software.^{11 19}

Feature definition

Features are recurring specific patterns in FOI image analysis that typically appear in several phases in certain diseases. The visibility of the features may be time-dependent; for example, some features appear shortly after ICG injection until they are superimposed by stronger superficial signals. Others are only visible towards the end of the image sequences. Twenty-six features were empirically defined in an atlas, which can be assigned to eight different groups: fingertip region (features I, Z, U, G, N, @, E), distal interphalangeal joint (DIP) region (features D and B), proximal interphalangeal (PIP) joint region (features P, X, A, O, T), metacarpophalangeal (MCP) joint region (features M and H), back of hands region (features J, W, V, S, F), wrist (features C, K, d), forearm (feature Y) and general behaviour (features r and R). All images were scored according to these features.

A summary of all 26 features including definition and respective FOI images can be found in online supplemental file 1. The typical phases of visibility of the patterns are indicated in the feature atlas.

Phase definition and feature reading

All patients underwent a 6 min FOI examination, resulting in 360 images per patient (one image per second). In comparison to images from the 3-phase model described in the literature,^{6 19} one finds significantly different appearance in frames around the signal maximum as well as in the late inflow phase. This observation led to the development of a new 5-phase separation of the image sequences, which is based on the 3-phase model as an extended procedure. For the latter, the overall time course of FOI in each hand was automatically divided into five phases using a computational algorithm. Using the length of the middle finger, a section line orthogonal to the forearm at the wrist is calculated for each hand. This defines the area of the hand. In this area, a computer algorithm calculates the 96th percentile of the signal intensity for the image sequence for each hand separately. Phases 1 and 2 describe the inflow (start to 15% of maximum intensity and 15%–90% on the rising slope, respectively), phase 3 is the peak phase, and phases 4 and 5 comprise the outflow (90%–36.8% and 36.8% to end on the falling slope, respectively). Phase images are the averaged images of the individual images of the image sequence of the respective phase. The images are scaled to the signal maximum in the hand area, a colour scale attributed and presented to the reader. The software provides a selection tool for features observed in the phase images. The user manually marks all features which he/she observes and may manually adjust the signal scaling so that the signals from the nail bed are presented in the range of maximum signal scaling and the signal scaling is not determined by small spots or signals from high intensity veins.

The 5-phase model is intended to reduce the susceptibility to errors by manual classification in the 3-phase model. It also enables a more precise and differentiated evaluation through more detailed phases and is standardised through automation. This is the first time it has been used, and it will be validated more thoroughly in the future.

The prima vista mode (PVM) comprises averaging the first 240 images. All phase images were evaluated on a yes/no basis by two readers (NS and JB) who were blinded to the diagnosis and calibrated against each other. Prior to phase 1 and during intensity adjustments at each subsequent phase, manual adjustments are necessary. These manual modifications have the capacity to exert an influence on both the level of agreement and the total scan time. A table showing the process step by step can be found in the appendix (online supplemental file 5).

Statistical analysis

The frequency of features for each diagnosis or group of diagnoses in each phase (PVM, 5-phase) was counted and statistically analysed. Contingency tables (in which the degree of freedom is one for each pattern and phase) were created with categorical variables counts and diagnosis. Using common formulae χ², diagnostic OR (DOR), true positive rate (TPR, sensitivity), true negative rate (TNR, specificity), positive predictive value (PPV) and negative predictive value were calculated. This test was chosen to show the degree of dependence of the occurrence of a feature for one of the diagnoses by rejecting the null hypothesis. We have developed a model in which OA is differentiated from the autoimmune diseases in the first step and the latter (RA, PsA and CTD) distinguished from each other in a second step. A p-value of <0.05 was considered as statistically significant. The inter-reader and intrareader reliability was calculated as not matching feature counts divided by the number of typical features per phase. Additionally, exact and close agreement rates (PEA: percentage of exact agreement, PCA: percentage of close agreement) were calculated.

Results

Clinical and demographic characteristics of the study population are shown in table 1.

Table 1. Demographic and clinical characteristics of the study population.

	RA (n=117)	PsA (n=110)	SLE (n=48)	SSc (n=62)	OA (n=90)
Disease duration, mean (IQR), years	2 (0.2–7.3)	1.1 (0–5.7)	10.1 (4.9–17.0)	5.5 (2.3–9.6)	9 (4–7)
Age, mean (SD), years	54.4 (12.5)	49.9 (11.2)	41.5 (11.9)	55.4 (18.7)	62.4 (9.8)
Sex, n (%), women	90 (76.9%)	73 (66.4%)	20 (95.2%)	33 (76.7%)	80 (88.9)
BMI, mean (SD), kg/m²	25.5 (4.2)	30.4 (0.4)	24.5 (4.5)	23.9 (3.4)	26.2 (4.2)
Use of disease-modifying anti-rheumatic drugs, n (%)	58 (48.7%)	27 (50.9%)	17 (80.9%)	–	–
Use of conventional synthetic disease-modifying antirheumatic drugs, n (%)	54 (45.4%)	20 (46.5%)	16 (76.2%)	–	–
Use of biological disease modifying antirheumatic drugs, n (%)	9 (7.6%)	9 (22.5%)	5 (23.8%)	–	–
Use of prednisolone, n (%)	80 (66.1%)	6 (15.4%)	11 (55.0%)	–	–
Disease Activity Score DAS-28, mean (SD)	5 (1.4)	4 (IQR 3.7–4.7)	–	–	–
Tender joint count, median (IQR)	6 (3–13)	4 (1–10)	4 (3–10)	–	8 (6–13)
Swollen joint count, median (IQR)	3 (1–6)	1 (0–3)	0 (0–0)	–	7 (4–10)
Patient global, VAS 0–100, mean (SD)	56.2 (21.4)	52.1 (21.9)	38.9 (22.1)	–	–

BMI, Body Mass Index; CTD, connective tissue diseases; OA, osteoarthritis; PsA, psoriatic arthritis; RA, rheumatoid arthritis; SLE, systemic lupus erythematosus; SSc, systemic sclerosis; VAS, Visual Analogue Scale.

We found that OA was best distinguished from the autoimmune diseases RA, PsA and CTD by the following features: In phase 1, patients with OA were less likely to demonstrate an early strong signal in the MCP joint (M), a signal from the area of the articular disk (K), or a retarded inflow of the dye (r), and in phase 2 by characteristics r or Raynaud’s syndrome (R). In phase 3, the presence of feature R was able to filter out an OA. In phase 4, patients with OA tended not to present with features M or an underperfused nailbed (U) and in phase 5 with feature U (see table 2).

Table 2. The differentiation between OA and the other diseases (RA, PsA and CTD).

		Other diseases vs OA
Features	Phases	c 2	DOR	TPR	TNR	PPV	NPV
M	1	19.1	3.50	0.21	0.93	0.92	0.23
K	1	3.6	2.34	0.07	0.97	0.90	0.21
r	1	8.7	2.82	0.13	0.95	0.91	0.22
R	2	9.35	3.03	0.12	0.96	0.91	0.22
r	2	8.65	4.15	0.09	0.98	0.94	0.22
R	3	12.0	8.24	0.08	0.99	0.97	0.22
U	4	8.10	2.49	0.14	0.94	0.90	0.22
M	4	8.16	2.59	0.13	0.94	0.90	0.22
U	5	7.75	2.94	0.11	0.96	0.91	0.22

CTD, connective tissue diseases; DOR, diagnostic OR; NPV, negative predictive value; OA, osteoarthritis; PPV, positive predictive value; PsA, psoriatic arthritis; RA, rheumatoid arthritis; TNR, true negative rate (specificity); TPR, true positive rate (sensitivity).

For the remaining autoimmune diseases RA, PsA and CTD, we defined two groups of features. The first stronger group had a TNR and PPV of >0.9 and the second less strong group had to have both values of at least 0.8 for TNR and PPV.

Patients with RA were more likely to present with an A-shaped pattern proximal from PIP (A) in phase 3. With TNR and PPV values above 0.8, RA was distinguished best if a cloudy pattern W occurs in phase 1, feature U in phase 2, or enhanced stripes DIP region and base of the finger (T) or by a broad signal enhancement pattern from DIP to PIP (B) in phase 3 (see table 3).

Table 3. PsA and CTD vs RA, RA and CTD vs PsA, RA and PsA vs CTD.

PsA and CTD vs RA
Features	Phases	c2	DOR	TPR	TNR	PPV	NPV
W	1	16.1	2.38	0.27	0.87	0.80	0.37
U	2	5.41	2.46	0.08	0.97	0.82	0.35
A	3	14.0	16.8	0.07	1.00	0.97	0.35
B	3	9.79	2.33	0.16	0.92	0.81	0.36
T	3	8.31	2.44	0.13	0.94	0.82	0.35
RA and CTD vs PsA
D	1	7.22	9.61	0.04	1.00	0.95	0.32
P	1	17.5	5.85	0.12	0.98	0.92	0.34
D	2	7.58	2.69	0.10	0.96	0.84	0.33
r	2	4.09	1.94	0.10	0.95	0.80	0.33
B	3	19.4	3.95	0.17	0.95	0.88	0.35
X	3	10.1	2.27	0.18	0.91	0.81	0.34
A	3	3.96	2.59	0.06	0.98	0.84	0.32
O	3	12.4	2.21	0.24	0.87	0.81	0.35
C	3	15.6	2.42	0.26	0.87	0.82	0.35
d	3	19.3	2.75	0.26	0.89	0.83	0.36
U	4	10.3	2.41	0.17	0.92	0.82	0.34
U	5	7.55	2.34	0.13	0.94	0.82	0.33
RA and PsA vs CTD
K	1	7.56	2.71	0.09	0.96	0.83	0.36
M	4	10.8	2.42	0.16	0.93	0.80	0.37
@	5	17.5	2.72	0.22	0.90	0.81	0.38
M	5	9.61	2.53	0.13	0.94	0.81	0.37

CTD, connective tissue diseases; DOR, diagnostic OR; NPV, negative predictive value; PPV, positive predictive value; PsA, psoriatic arthritis; RA, rheumatoid arthritis; TNR, true negative rate; TPR, true positive rate.

Patients with PsA were more likely to show signal enhancements above DIP (D) or PIP (P) joint regions in phase 1. Furthermore, the features D or r (in phase 2), the features B; X; A; O; C, d (in phase 3) and U (in phases 4 or 5) were able to distinguish PsA from CTD and RA (see table 3).

Patients with CTD presented various less strong features: feature K in phase 1, M in phase 4 and outgoing lateral strand (@) or M in phase 5 (see table 3).

Inter-reader reliability across all phases was moderate (0.61–0.82). The agreement rates were highest for PsA (0.69–0.82) and lowest for CTDs (0.61–0.72). Generally, the first three phases showed higher agreements (0.62–0.82) than phases 4, 5 and PVM (0.61–0.76). The intrareader reliability was in the range of 0.79–0.86 for reader 1 (NS) and 0.74–0.83 for reader 2 (JB). The PEA was lowest in phase 4 compared with all other phases (see onlinesupplemental files 2 4).

Safety

All patients tolerated the examination well. Allergic reactions, hypersensitivity reactions or other adverse conditions were not observed.

Discussion

The aim of this analysis was to differentiate RA, PsA, CTD and OA from each other by using recurring features in FOI to improve the diagnostic process for the early and most accurate diagnosis of typical rheumatic diseases affecting the hands, which is especially relevant in the early arthritis clinic.

Previous studies have extensively investigated the use of FOI for the detection of inflammation in RA,^{7 8} PsA,⁹ CTDs^{10 11} and OA^{12 13} by signal enhancement of ICG in the affected areas,¹⁸ mainly in the joints. In these studies, the authors focused on inflammation detection by the use of FOIAS, but did not perform feature analysis. Moreover, only one comparative study of OA versus RA exists so far, in which also a focus on joint inflammation was performed.¹² Another previous study has examined feature findings in FOI for RA, OA and CTD.²⁰ However, they focused on extraction of relevant features and did not evaluate in detail the diagnostic value of the feature analysis and only evaluated cases with a smaller feature catalogue limited to the examined diseases in a 3-phase model with one fixed phase.²⁰ Thus, a direct comparison of the previous studies with our current results is difficult to perform.

In a stepwise procedure, which is presented here, we first wanted to distinguish and, by that, exclude OA from the autoimmune rheumatic diseases RA, PsA and CTD by FOI. However, we did not find single features that have high sensitivity and specificity at the same time, which must be the properties of a differentiator. Instead, several features with high specificity were identified that reject patients from belonging to a group with a specific disease. In detail, we found that the features R, r, M, K and U should raise the suspicion of RA, PsA or CTD rather than OA.

In a second step, RA was differentiated best from PsA and CTD by the five different features: A, W, U, B and T.

Rather than studying joint inflammation, skin involvement or therapeutic monitoring like in previous studies, we focused on the use of FOI for differential diagnosis. Werner et al have defined a PsA-specific feature in a previous study, and later, the ‘green/blue nail’ phenomenon was defined by Wiemann et al.¹⁷ In the presented analysis, we were not able to identify specific PsA features, and in order to distinguish PsA from RA and CTD, even the highest number of features (in total, 12) had to be used, as follows: inflamed DIP region (D), circular signs from inflamed joint (P), diffuse dorsal carpal ligament sign (d), broad signal from DIP to PIP (B), A-shaped pattern proximal from PIP (A), O-shaped signal around the PIP joint (O) and early signals in the region of carpal joints (C). This reflects the heterogeneity of PsA, which affects the joint and periarticular structures (tendons and enthesitis) as well as the skin. In a detailed analysis, the separate calculation of SLE and SSc data did not give better results. This confirms our hypothesis that a separate consideration of the corresponding diseases in the context of this analysis is not expedient. Both are CTDs with main symptoms of Raynaud’s phenomenon. In the early stages of the disease, the clinical and imaging features can be very similar and an overlapping appearance is not uncommon. This close relationship is also reflected in the tabular summaries (online supplemental file 3). A more distinguished analysis of CTD subgroups was not the aim of the study; rather, we focused on the differentiation of disease groups (OA, PsA, RA and CTD) on the basis of imaging features.

In general, rheumatic diseases are often complex, which accounts for the observed inhomogeneity among patient groups. A low TPR indicates that the pattern is not present in all patients, but a high PPV suggests that when the pattern is present, the patient likely belongs to the cohort. However, the absence of the pattern results in a poor DOR. This variability may stem from including patients from different studies, although stricter inclusion criteria would not accurately reflect real-world clinical practice.

In the past, several studies have shown that FOI is a useful tool for also investigating reduced microcirculation and vasculopathy in SSc.^{15 21 22} To date, there are no published studies yet investigating the role of FOI in SLE or other CTDs (eg, Sjögren), especially in terms of disturbed microcirculation in joints or skin of the hands. In the current analysis, we found that a shepherd’s crook pattern (laterally outgoing strand, @), an early strong signal between MCP joint (M), and a signal from the area of the articular disk (K) proved to exclude CTDs from RA and PsA. Here, mostly FOI images of SSc (n=62) and SLE (n=48) patients were included. In the future, criteria should be found to not only differentiate CTDs from inflammatory joint diseases, but also from different forms of CTDs (eg, SLE vs SSc).

Rothe et al identified most valuable features for differentiation in another phase model. However, the results cannot be used directly in clinical practice. With our work, we confirm their outcome for a different phase model that only a limited feature set is necessary for differential diagnosis.

To the best of our knowledge, this is the first study to use FOI as a tool to distinguish between different rheumatic diseases affecting the hands using feature analysis. We were able to show that the diagnosis of both inflammatory rheumatic diseases (RA, PsA and CTD) and non-autoimmune disease OA can be improved by using a stepwise process. In a first step, OA is ruled out based on various (five) features, and afterwards, patients with an inflammatory rheumatic disease can be referred to the early arthritis during the ‘window of opportunity’. Moreover, CTDs often present organ involvement, which helps to reach a faster diagnosis and treatment through this process. Therefore, the ultimate aim of this analysis would be to implement FOI through a new diagnostic pathway in the early arthritis clinic or even before seeing a rheumatologist, that means already at the general practitioner (GP) or the dermatologist. In the first step, OA is being ruled out and in the second step, FOI feature reading is used to differentiate between the typical autoimmune rheumatic diseases RA, PsA and CTD already at the rheumatology clinic (usually, early arthritis clinic). At this stage, other parameters such as clinical findings and laboratory parameters are integrated into the diagnostic process anyway, as well. The somewhat poorer statistical values here can be explained by the fact that they, as inflammatory diseases, are more challenging to differentiate. But at this (second) step, FOI gives a clear indication of the direction in which the diagnosis should be pursued, and which further examination would be useful.

In the future, rheumatologists will no longer have to go through all the FOI features and phases, but will only have to look for certain combinations of features in certain phases, or this will be performed by automated pattern recognition algorithms through artificial intelligence.

We are aware of some limitations of our study. First, due to the retrospective study design, we had to deal with a few incomplete data. Second, the inter-reader reliability was only moderate and therefore needs to be improved. As with all imaging techniques, the results of FOI image analysis depend on the experience and routine of the operator. In the future, to improve the objectivity of the ratings, efforts should be made to increase the inter-reader reliability, especially in phases 4 and 5. We are confident that with increased use and experience of FOI, and more tightly defined features, inter-reader reliability will be improved.

In the current study, patients were divided into subgroups according to disease only. No subgroups were formed regarding disease duration and possible current antirheumatic (especially glucocorticoid) therapy affecting inflammatory joint or skin involvement. It is therefore important that future studies investigate whether and to what extent patients’ disease stage, disease duration and current anti-inflammatory therapies influence the feature analysis.

The statistical analysis employed in this study utilises contingency tables comprising two categorical variables to distinguish diseases or groups of diseases. However, combinations of observed and non-visible patterns at different phases will further improve differentiation. These combinations from all phases are best determined using machine learning algorithms. Due to manual adjustment, the scoring is semiquantitative. In future studies, the implementation of a fully automated system is recommended, as it has the potential to reduce variability and improve agreement.

A major strength of FOI imaging is that it is easy to use and well tolerated. In addition, the patient is not exposed to any radiation during the examination. Unlike ultrasound, for example, FOI can be performed by a medical assistant, as the examination does not require the presence of a doctor. Evaluation of the images by a doctor can also take place later, allowing for remote working in the future and greatly increasing flexibility. However, it should be noted that FOI is not yet a procedure that can be fully transferred to medical assistants, as the evaluation must always be conducted by a trained medical doctor. Although FOI is an invasive imaging technique that requires an injection of ICG, studies show that it is very well tolerated by patients and side effects such as allergic reactions are very rare (1:42000).^{23 24} As the amount of injection is variable, FOI can also be used on minors.²⁵ In addition, the FOI examination of all fingers can be described as a relatively short examination with an examination time of only 6 min. FOI can be considered a safe, rapid and low-threshold examination method to display inflammation in the hands, pathognomonic changes and skin involvement.¹⁹ A limitation of FOI is its invasiveness, which must be tolerated by the patient, and its inability to visualise bone destruction anatomical structures, unlike X-rays. However, this is of secondary importance in modern rheumatological diagnostics, as early diagnosis of RA mostly occurs before significant bone damage is present, especially at the initial diagnosis stage when FOI is most commonly used. In the future, these factors may help to improve patient care, especially in rural areas with a low density of rheumatologists, leading to an earlier and more accurate diagnosis and initiation of therapy through the presented stepwise process. Despite this, the 5-phase model used in this study is a new procedure that has not yet been validated, and further studies are needed.

The comparatively low PCA agreements suggest that the methodology needs to be optimised and that the readers require improved training. Differentiating between OA and RA is particularly clinically relevant, and the PCA values currently provide the best results, although there is still room for improvement. PEA/PCA data shows higher agreement in phases 1–2 and PVM (20%–40%) than in phases 3–5, where agreement is particularly low. This suggests that individual assessment is highly dependent. To achieve more consistent results, standardised training protocols and clear assessment criteria must be developed and implemented.

In conclusion, we presented the first atlas of individual FOI patterns for feature analysis, improving differentiation of typical rheumatic diseases affecting the hand. This may simplify the diagnostic process in the future. Thus, FOI should be viewed as an adjunct, ancillary component of a broader imaging repertoire, not a standalone modality. In the future, it could support GPs in areas lacking specialists and may help triage patients for rheumatology referral. Specialists can then follow-up with, for example, ultrasound, MRI and other relevant examinations. Its high PPV aids in diagnostic uncertainty and may be particularly relevant in cases of diagnostic uncertainty. Further, in cases of differentiating OA and seronegative arthritis, where other modalities may be less definitive, FOI could make an impact.

The need for such a tool is explained by the fact that early, accurate diagnosis and, above all, the earliest possible start of therapy can significantly improve patient outcomes.^{26 27} This is even more important since the number of specialists in rheumatology will be reduced during the next years, especially in rural regions. The widespread use of FOI could therefore contribute to an improvement in medical care in rheumatology.

Data availability statement

Data are available upon reasonable request.