An objective, automated and robust scoring using fluorescence optical imaging to evaluate changes in micro-vascularisation indicating early arthritis.

Fluorescence optical imaging technique (FOI) is a well-established and valid method for visualization of changes in micro vascularization at different organ systems. As increased vascularization is an early feature of joint inflammation, FOI is a promising method to assess arthritis of the hands. But usability of the method is limited to the assessors experience as the measurement of FOI is semi-quantitative using an individual grading system such as the fluorescence optical imaging activity score (FOIAS). The goal of the study was to automatically and thus, objectively analyze the measured fluorescence intensity generated by FOI to evaluate the amount of inflammation of each of the subject's joints focusing on the distinction between normal joint status or arthritis in psoriatic arthritis patients compared to healthy volunteers. Due to the heterogeneity of the pathophysiological perfusion of the hands, a method to overcome the absoluteness of the data by extracting heatmaps out of the image stacks is developed. To calculate a heatmap for one patient, firstly the time series for each pixel is extracted, which is then represented by a feature value. Secondly, all feature values are clustered. The calculated cluster values represent the relativity between the different pixels and enable a comparison of multiple patients. As a metric to quantify the conspicuousness of a joint a score is calculated based on the extracted cluster values. These steps are repeated for a total number of three features. With this method a tendency towards a classification into unaffected and inflamed joints can be achieved. However, further research is necessary to transform the tendency into a robust classification model.

Introduction

Musculoskeletal (msk) pain, especially joint pain, is a very common symptom potentially leading to chronic pain syndrome and impact of function if a proper treatment is not initiated. Mechanical pain is the most common reason for joint complains, but pain can also be caused by inflammation associated to immune mediated diseases. Different types of arthritis can be identified, from which rheumatoid arthritis is the most frequent one, affecting approximately 1% of the adult European population [1]. The differentiation between the mechanical and inflammatory cause for joint pain is of high importance to avoid structural damage of the joints by choosing a specific treatment.

In clinical practice, arthritis is diagnosed by clinical examination of swollen and tender joints by rheumatologists. Additionally, imaging techniques can be used to quantify the amount of inflammation. Imaging methods such as ultrasound with power-doppler technique are widely used in Europe but limited to the assessor’s experience, high inter-reader variability and to its time consuming procedure when used for assessment of all joints of the msk-system. MRI can be used as an alternative to ultrasound with a higher sensitivity and specificity, but has limitations due to its accessibility and expensiveness. In early stages of arthritis, swelling may not be evident whereas inflammation can be detected sensitively using imaging methods of the joints (e.g. ultrasound or MRI). Fluorescence-optical imaging is an indocyanine green (ICG) tailored method to visualize micro vascularization of the hands [2-4]. It might be of high value for detection of early arthritis when clinical examination may not lead to a clear diagnosis as it is well-tolerated by the patients and easily accessible for the physicians. By now, FOIAS is assessed semi-quantitatively and depends on the reader’s experience in evaluation of the film [2-4] although fluorescence intensities are measured quantitatively.

The aim of this work is to overcome the semi-quantitative assessment method and replace it with an objective, reproduceable and quantitative assessment system for the FOI images. Comparing the per se signal between subjects is not trivial, since patients have very heterogeneous perfusion of the hands. Therefore, defining threshold values describing the health status of each joint, using the raw data, is impossible. The suggested method overcomes the absoluteness of the measured data by calculating heatmaps visualizing conspicuous pixels regarding different feature values applied in a cohort of patients with psoriatic arthritis compared to healthy volunteers.

Related work and comparison to the proposed method

Previously the value of fluorescence optical imaging (FOI) as a diagnostic tool to detect any kind of joint inflammation e.g. synovitis was shown [5, 6]. Furthermore its specificity and sensitivity in comparison to 1.5 T MRI [7], 3 T MRI [8], ultrasonography in grey scale and power Doppler mode [6, 7] and clinical examination [6-8] were analysed and described in detail. Additionally, there were attempts to automatically classify patients as well as single joints into healthy and inflamed [9, 10]. In [9] the time series signal is extracted from the joint areas and the health status distinguished by the time series. Even though Dziekan et al. achieved a distinction between affected and unaffected joints within one patient diagnosed with rheumatoid arthritis, this approach shows a lack of comparability between patients. This incomparability is caused by physiological variability and inter-individual differences of the characteristics of the microvascularisation. In this work heatmaps based on three features extracted from the image stack are generated to overcome this challenge. In [10] a principal component analyses (PCA) over temporal subsequences of the image stack are performed. With this approach a promising result was achieved to distinguish between “healthy or mild synovitis and moderate or severe synovitis” [10, p.14]. However, a lack of distinction between unaffected and mild synovitis was observed. Nevertheless, there is a high medical need for distinguishing between not and mild inflamed joins.

Patient group

All included patients were diagnosed with active psoriatic arthritis (PsA). Additionally, 12 healthy volunteers without any joint complains were recruited and are used as a control.

For 163 patients an assessment of clinical signs of PsA (swollen and tender joint count and its evaluation by a rheumatologist) is available.

FOIAS was measured in 91 patients using grading according to ICG-distribution [2-4]. Increase of vascularization in the different ICG-distribution phases was rated by an experienced central reader on a 0–3 scale. Here, a score of 0 represents no visible conspicuousness, while score 3 indicates high visibility.

The 12 healthy volunteers were selected with regards to the following in- and exclusion criteria: (a) missing musculoskeletal complaints on the hands, (b) missing diagnosis of joint diseases, (c) missing comorbidities with focus on diseases that go ahead with joint diseases, (d) missing contraindications to use indocyanine green as colour agent for FOI examinations. Due to these criteria, the 12 volunteers are not affected by PsA and can act as a control group.

For an unbiased investigation of the proposed method, no further subgroups are formed. Data were analyzed in a blinded manner. Therefore, all additional parameters, which can influence the perfusion of the hands e.g. gender, age, weight, smoking habit, surrounding temperature etc. are not taken into account.

For performance of the analysis of the pseudonymized clinical and imaging data, patients / volunteers were recruited from two prospective non-interventional studies (one with inclusion of patients diagnosed with psoriasis or psoriatic arthritis and one with inclusion of healthy volunteers without complaints in the joints of the hands). The study fulfilled Good Clinical Practice Guidelines and all patients / volunteers provided signed informed consent for inclusion in the studies and agreed to the usage of their data for research purposes. All clinical procedures were performed according to study protocol, which received ethical approval from the ethics committee of the University Hospital Frankfurt a. Main, Germany. All patients / volunteers were fully capable to give informed consent for participation in a study.

Data acquisition

The proband places her/his hands into a preformed template beneath the camera and inside the Xiralite X4 machine. Then, a run to acquire the images is started. To reduce the influence of the environment on the data, the room’s windows are covered and the lights are switched off. After 10 s the prepared colour agent indocyanine green (ICG) is injected into one of the proband’s arms with a dose of 0.1 mgICG/kgbody weight. The measurement takes six minutes in which an image stack of 360 pictures is taken (one picture per second). The colour agent is the fluorescent substance and thus essential for each measurement. Fig 1 visualizes the measured data.

Fig 1

Image stack (for visualization purposes).

The red line visualizes the extraction of one pixel time series.

Method

All described methods were implemented using Python.

To calculate the score for each joint in the hands, two separate main processes have to be carried out and combined to a final result. Firstly, heatmaps based on features extracted from the image stack are generated. Secondly, the joints’ positions and sizes are determined. Finally, the heatmaps are investigated and evaluated in the defined joint areas. However, this work mainly focuses on the heatmap calculation and final evaluation. Thus, the segmentation of the joint areas is only described briefly.

Binary mask extraction

Since a binary mask is used in the segmentation as well as the heatmap calculation (section Joint Segmentation and Generating the heatmaps based on the image stack), the mask extraction from the image stack is crucial for the final result. To calculate this mask a maximum image is generated based on the 360 image time series. Each pixel of the maximum image corresponds to the highest value of the according pixel time series. After enhancing the contrast of the maximum image, a bounding box around the two hands is defined and the “GrabCut” algorithm [11] is used to get the binary mask. The bounding box is defined by the spiking pixel values within the hand region. The different steps are visualized in Fig 2A–2C.

Fig 2

Steps of the binary mask extraction and result of the joint region calculation.

(A) Extracted maximum image. (B) Enhanced image with bounding box. (C) Final binary mask. (D) Final joint areas.

Using the method with the maximum image guarantees that the final mask includes possible movement artefacts. The tracking of the exact movement is lost, however the proposed heatmap approach is robust against smaller movements.

Joint segmentation

Two methods are combined to determine the joints’ positions, since none of them performs well on all joints. One algorithm is based on classic segmentation techniques combined with anatomical hand proportions [12]. In this approach, based on the extracted binary mask (see section Binary mask extraction), the fingers are detected and their length calculated. Since previous work have set the joints’ locations in proportion to their finger and the corresponding finger length, the position can be estimated. This method works best for the Proximal (PIP) and Distal (DIP) Interphalangeal joints.

In the second method the joints’ locations are calculated with the pretrained neural network OpenPose [13]. With this approach all joint positions are calculated at once. However, only the precision for the Metacarpophalangeal (MCP) and Interphalangeal (IP) joints increases in comparison to the classical approach and therefore, are used for further analysis.

All joint regions are defined as circles. The radius is either defined over the smallest distance between the defined joint location and the finger edge (IP, DIP, PIP) or the smallest distance to the next joint location (MCP II-V). The radius of the joint MCP I is equal to the diameter of the thumb (the result is visualized in Fig 2D).

Generating the heatmaps based on the image stack

The time series has a characteristic shape as described in [9]: a steep ascent, followed by a slow descent after reaching the peak (compare Fig 3A). Even though Fig 3A visualizes the average pixel value of one image at certain time points, each pixel within the signal region (hand region) follows a similar shape.

Fig 3

Smoothing and feature extraction of one representative pixel time series.

(A) Gaussian filtering of the pixel time series. (B) Feature extraction of the pixel time series.

In the proposed approach, the time series is extracted at each pixel (an example is shown in Fig 3). Each time series is used to extract three features. Thus, each pixel is represented by three independent values. All values of one feature are collected in one set of data. Therefore, three independent sets of data are created, in which each value represents one pixel. The three features are the amplitude ΔI = Imax − Imin, the mean pixel value during the increasing time (a denoting the frame for I and b the frame for I) and the maximal gradient during the increasing time max{I − I} with i ∈ {min ≤ i ≤ max −1} (compare Fig 3B). Since, the data is discrete and thus the difference between the current and the next point is calculated. The time point in which the calculated difference has the highest value is defined as maximal gradient.

To create the heatmaps from the three data sets, k-means clustering is performed on each of the data sets. Therefore, every pixel gets a cluster assigned, which is represented by a colour. The corresponding pixels in the (still empty) heatmap is set to the assigned pixel color. For k-means clustering the number of centroids has to be predefined. Since the best number of centroids is unknown, k-means clustering is performed four times choosing k as 3, 5, 7 and 9. Thus, for each of the three features 4 heatmaps are created (an example is shown in Fig 4).

Fig 4

Calculated heatmaps for the three features.

The row specifies the feature and the column the amount of cluster used creating the heatmap. For this example the slope feature (third row) emphasise the different health conditions of the joints. In the slope three cluster heatmap the right DIP II joint is assigned to the same cluster as the left PIP III joint. The slope seven and slope nine cluster heatmap represent the actual inflammatory condition better and show a clear distinction between right DIP II and left PIP III joints. For different examples each feature shows a different classification power. (A) 3 cluster amplitude. (B) 5 cluster amplitude. (C) 7 cluster amplitude. (D) 9 cluster amplitude. (E) 3 cluster mean. (F) 5 cluster mean. (G) 7 cluster mean. (H) 9 cluster mean. (I) 3 cluster max slope. (J) 5 cluster max slope. (K) 7 cluster max slope. (L) 9 cluster max slope.

Additional steps to lower the noise in the heatmaps

In the process of generating the heatmaps two additional steps are performed to decrease the noise in the heatmaps. Firstly, to suppress any kind of interference with the background signal, the binary hand masks (see section Binary mask extraction) are applied and the background is completely set to 0. Secondly, after extracting the time series for one pixel, the data is smoothed by applying a Gaussian filter (standard deviation σ = 1). Therefore, outliers have less impact on the outcome (compare Fig 3A).

Heatmap interpretation

Fluorescence optical imaging visualizes the current distribution of the colour agent ICG. Joints suffering from an inflammation show a higher perfusion [14] with new formation of vessels especially in PsA. The assumption is that inflamed joints have a higher signal in comparison to non-inflamed hand regions. Due to the higher perfusion into the joints the signal increases faster than in non-inflamed hand regions.

The three features are chosen to represent these properties of the perfusion. The amplitude and mean feature correspond to the amount of blood and the slope to the streaming speed. Even though the amplitude and mean represent the same physical property, the mean feature includes more of the time dependency of the data. Especially, fluctuations or a decline of the slope is not affecting the amplitude value.

Scoring the joints based on the heatmaps

In the final step the heatmaps are investigated and evaluated within the defined joint areas (compare Fig 5). At first a value is assigned to each pixel within the joint area depending on its color. The lowest cluster (blue) (compare any picture in Fig 4) is represented by 0, while the highest cluster (red) is represented by the number of clusters minus one. All clusters in between get the corresponding number assigned. For example in a seven cluster heatmap blue pixels add 0 to the joint score, while red pixels add 6 to the joint score. To calculate the final score S for one joint j, all assigned pixel values p are added and divided by the number of pixels n and the number of clusters m.

Fig 5

Example for combining the extracted joint (compare Fig 2D) areas with a heatmap (compare Fig 4K).

Due to the new idea of evaluating the data, the calculated scores do not have an explanatory power without setting them into context. However, due to the explanations in section Heatmap interpretation the assumption is made that for all features the scores for unaffected labelled joints are in general lower than for affected labelled joints.

Result

To evaluate the proposed method’s eligibility to detect affected joints the calculated scores are connected to two different labels: the clinical label (swollen and tender joint assessed by clinical examination) and the FOIAS (described in section Patient group). The two different labels include three and four sub-categories.

clinical label with the sub-categories: affected (swollen or tender), swollen or tender

FOIAS with the sub-categories: affected (score higher 0), score = 1, score = 2 or score = 3

To visualize the statistical outcome of all sub-categories notched box plots are used. Each sub-category contains 9 box plots, which form 3 groups of 3. Each group visualizes the calculated score distribution for one feature (amplitude, mean and slope). In each group the left (red) box plot represents the scores for affected (PsA) and the box plot located in the middle (blue) the unaffected labelled joints for patients with a confirmed PsA. The third box plot on the right (purple) describes the distribution for healthy volunteers. Additionally, the mean value for each box plot is added as a black dot.

However, the results for the different amounts of cluster are not integrated into this visualization. Therefore, each figure not only represents one sub-category but also the amount of cluster (e.g. sub-category: swollen, cluster: 7). For clarity only the figures visualizing the results for k = 7 are embbeded into the manuscript. The remaining graphs for all other clusters as well as the mean and median values are visualized and summarized in S1–S6 Figs.

Finally, the predictive power of trained machine learning systems based on the three features and labelled by the clinical label as well as the FOIAS are investigated.

Scoring results in comparison to the clinical labels

In section Patient group it is mentioned that for 163 patients with a confirmed PsA a clinical assessment (swollen or tender joints) is available, which correspond to 163 ⋅ 28 = 4564 assessed joints. 3824 joints were labelled as unaffected, 455 as tender and 285 as swollen. Furthermore, the 12 healthy volunteers result in 336 healthy labelled joints.

In Fig 6 the results for the different sub-categories of the clinical label are summarized. The three graphs including the 9 box plots show all similar distributions of the calculated joint scores. Even though there is a big overlap between the unaffected and affected (tender, swollen, tender and swollen) distributions, a clear tendency towards a lower score in unaffected labelled joints can be observed. This observation is supported by the median and average scores. For 33 out of 36 groups (four cluster, three feature, three label) the notches comparing the distributions of unaffected and affected labelled joints do not overlap (Fig 6 and section S2 Fig). This suggests that the true medians of these distributions differ with a confidence of 95% [15]. The scores for the healthy volunteers are in general lower than the affected and unaffected score distributions.

Fig 6

Comparison between the calculated scores for unaffected and affected labelled joints based on clinical labelling (k = 7).

(A) Comparison between affected and unaffected labelled joints (sub-category: affected, k = 7). (B) Comparison between tender and unaffected labelled joints (sub-category: tender, k = 7). (C) Comparison between swollen and unaffected labelled joints (sub-category: swollen, k = 7).

However, for the tender labelled joints the notches of the healthy distribution overlaps with the notches of the unaffected and affected distributions for the slope feature choosing k as 3, 7 or 9. Additionally, the healthy probands show an overall smaller scattering than the other two distributions.

For the swollen labelled joints no overlap between the healyth distribution and the swollen distribution among all clusters and features can be observed.

The three different sub-categories show different abilities to classify the joints. The distribution for tender joints (Fig 6B) shows the smallest and the swollen distribution (Fig 6C) the biggest difference between unaffected and affected joints. Since Fig 6A includes both, tender and swollen joints, it represents the average distribution.

Scoring results in comparison to the experts evaluation of the FOI images performed by an experienced central reader using FOIAS [2–4]

For all calculations the maximal joint FOIAS score among all three phases is considered as the label value.

In section Patient group it is mentioned that for 91 patients a FOIAS assessment is available, which correspond to 91 ⋅ 28 = 2548 assessed joints. 2278 joints are labelled as unaffected, 102 as score = 1, 150 as score = 2 and 18 as score = 3. The healthy distribution is again formed from the 12 healthy volunteers.

In Fig 7 the results for the different sub-categories of the FOIAS are summarized. The graphs show, that with increasing FOIAS score the overlap between unaffected and affected labelled joints decreases and thus, the distinction between unaffected and affected increases (compare Fig 7–7D). The same tendency can be observed by comparing the medians and means of the distributions. Furthermore, besides a slight overlap for the notches of the slope feature in the score = 1 graph, no overlap between the affected and unaffected distribution can be observed. This suggests that the true medians differ with a confidence of 95%. The score distributions for the healthy volunteers are in general lower than the other two distributions. However, similar to the clinical assessment the notches of the unaffected and healthy distributions overlap for the slope feature. For some of the sub-category score = 1 the notches of the affected and healthy distribution overlap as well. Due to the small sample size of score = 3 labelled patients, in same cases the confidence interval of the median extends the third quartile, which results in an unusual boxplot shape.

Fig 7

Comparison between the calculated scores for unaffected and affected labelled joints using FOIAS (k = 7).

(A) Comparison between affected and unaffected labelled joints (sub-category: affected, k = 7). (B) Comparison between score = 1 and unaffected labelled joints (sub-category: score = 1, k = 7). (C) Comparison between score = 2 and unaffected labelled joints (sub-category: score = 2, k = 7). (D) Comparison between score = 3 and unaffected labelled joints (sub-category: score = 3, k = 7).

Fig 7A includes all joint scores with a FOIAS score higher than 0 and thus, represent the overall possibility of the suggested method to distinguish between unaffected and affected.

Investigation of a trained machine learning system to predict the health status of a joint based on the calculated features

In order to investigate the predictive power of the three scores calculated for each joint to predict the heath status a random forest [16] classifier for the k = 7 scores is trained. The performance is independently investigated for the two different labelling methods (clinical labelling and FOIAS labelling). Since, both cases suffer from class imbalance the sizes of the overrepresented classes are cropped to the smallest class size. For the clinical label the class with the least amount of examples contains 740 data points and for the FOIAS label 18 data points. For both cases the data points of the overrepresented classes are picked randomly. Due to the cropping many data points remain unused. Therefore, the classifier has been trained 500 times for each labelling method.

Since for the clinical labelling a joint can be labelled as unaffected, swollen, tender or swollen and tender the classifier has been trained to distinguish only between unaffected or affected scores. Thus, a binary classification problem has to be solved. The calculated F1 scores range from around 40% to around 62% detection rate. The results are visualized in Fig 8A.

Fig 8

F1 score results of the trained and tested random forest classifier using scores calculated with k = 7 clusters.

(A) Results for the 500 runs using the clinical label as ground truth. (B) Results for the 500 runs using the FIOAS label as ground truth.

For the FOIAS labelling a 4 class (Score 0, 1, 2 or 3) classification problem has to be solved. The calculated F1 scores range from 0% to 66% detection rate. The results are visualized in Fig 8B.

Discussion and conclusion

The calculated scores show the expected outcome. The scores for unaffected labelled joints are in general lower than the scores for joints labelled as affected. Furthermore, the distributions for tender joints show the smallest difference between unaffected and affected joints. This can be explained with the lowest comparability between patients, since the sense of pain is highly subjective. Additionally, the assumption is made that a swelling goes along with an inflammation in the affected joint, which would lead due to the increased perfusion to a higher signal (compare section Heatmap interpretation). Since more joints are labelled as tender than labelled as swollen the average distribution (compare Fig 6A) is closer to the tender distribution (compare Fig 6B) than the swollen distribution (compare Fig 6C). Furthermore, the distributions for the FOIAS label shown in Fig 7 correspond to the expected outcome, since a higher FOIAS score is equal to a higher visibility in the images for the investigated joint.

Across all labelling methods, the score calculations based on the slope feature seem to lead to a higher overlap between unaffected and affected labelled joints. One explanation for the overlap is that in theory the finger tips and fingers are firstly visible in the image stack [17]. Therefore, k-means clustering will classify these regions as conspicuous, regardless of the perfusion of any other patient. This affect of the relative data is most dominant in the slope feature, since the blood streaming is not homogeneously distributed over the hand region. Furthermore, 64% of the joints are located within fingers, which emphasises the anatomical bias.

Wider notches of some distributions in comparison to other distributions, which in a few cases leads to an overlap of the notches, are caused by the smaller sample set.

Comparing the different numbers of clusters leads to the conclusion, that a higher number does not necessarily lead to a clearer distinction between the unaffected and affected score distribution. However, comparing the box sizes of the 3 cluster and 9 cluster box plots suggest that the variety within the distribution based on the 3 cluster heatmap is higher than the distribution based on the 9 cluster heatmap. Thus, choosing k as 7 or 9 seems sufficient. The observed results indicate, that the presented idea can overcome the high inter-patient variability in the data, hence the majority follows the same tendency (compare Figs 6 and 7, S1–S6 Figs).

The approach to train a machine learning system based on the calculated feature scores does not lead to a robust classifier. The fluctuating results can be explained by the combination of overlapping scores and class imbalances as well as the features’ properties. Due to the cropping and randomly picking of data points, the overlap of scores differs for each run, which leads to fluctuating results. Additionally, the data shows that an affected joint can have a high variety of score combinations along the different features. Therefore, an understanding of the causes for the heterogeneity within the data, the impact of non-disease related factors on the data and developing prepossessing steps to homogenise the data is crucial.

The comparison between unaffected and FOIAS score = 1 labelled joints is the most interesting, since this work is embedded into a project with the goal of the early detection of arthritis. Even though a high overlap between unaffected and affected labelled joints can be observed in Fig 7B, it also shows a clear tendency towards a classification into missing or mild arthritis. Therefore, the suggested objective, reproduceable and quantitative assessment system shows a promising first result and motivates further analysis.

Future work

The presented method could lead to a sufficient diagnosis of arthritides such as PsA. However, there are limitations planned to be addressed in a following study.

Non-disease related impact factors

To investigate the non-disease related impact factors and therefore, understanding the heterogeneity in more detail, a study among heathy volunteers is planned. The impact of factors like hand temperature, BMI, alcohol consumption and many more are planned to be addressed.

Normalization of the images on the used machine

In the proposed approach the features values of all pixels in the image are compared with one another, regardless of their location within the image. However, the illumination of the Xiralite device is not homogeneous and thus, the pixels are technically incommensurable. The centre part of the image is in average brighter than the outer part. Therefore, taking a reference background image, capturing the light gradient in the device, and applying it to the raw images leads to a commensurable data set.

Automated feature extraction

The heatmaps are calculated based on the extracted three time series features amplitude, mean during the increasing time and maximal slope. Even though, these three features already enabled to achieve a tendency for unaffected and affected scores, there could be features with a stronger classification power. With enough data a machine learning approach could choose better features and lead to better results.

Normalization of the images on the proband

With the heatmap approach a method to overcome the heterogeneity of the data and to achieve an inter-proband comparability is suggested. However, only focussing on relative data can lead to unexpected outcomes. For example a completely healthy person could get in average a medium high score due to fluctuation within the data, even though all joints should lead to low scores. This phenomenon is already observable for the healthy probands. A process combining the absolute data with the relative data could lead to much clearer results.

Definition of analysed joint areas

The conspicuous areas and the joint areas do not always match. Therefore, a method is needed to evaluate the heatmaps at the correct areas.

Comparison between the calculated scores for unaffected and affected labelled joints based on clinical labelling (k = 3).

(A) Comparison between affected and unaffected labelled joints (sub-category: affected, k = 3). (B) Comparison between tender and unaffected labelled joints (sub-category: tender, k = 3). (C) Comparison between swollen and unaffected labelled joints (sub-category: swollen, k = 3).

(TIF)

Click here for additional data file.

Comparison between the calculated scores for unaffected and affected labelled joints based on clinical labelling (k = 5).

(A) Comparison between affected and unaffected labelled joints (sub-category: affected, k = 5). (B) Comparison between tender and unaffected labelled joints (sub-category: tender, k = 5). (C) Comparison between swollen and unaffected labelled joints (sub-category: swollen, k = 5).

(TIF)

Click here for additional data file.

Comparison between the calculated scores for unaffected and affected labelled joints based on clinical labelling (k = 9).

(A) Comparison between affected and unaffected labelled joints (sub-category: affected, k = 9). (B) Comparison between tender and unaffected labelled joints (sub-category: tender, k = 9). (C) Comparison between swollen and unaffected labelled joints (sub-category: swollen, k = 9).

(TIF)

Click here for additional data file.

Comparison between the calculated scores for unaffected and affected labelled joints using FOIAS (k = 3).

(A) Comparison between affected and unaffected labelled joints (sub-category: affected, k = 3). (B) Comparison between score = 1 and unaffected labelled joints (sub-category: score = 1, k = 3). (C) Comparison between score = 2 and unaffected labelled joints (sub-category: score = 2, k = 3). (D) Comparison between score = 3 and unaffected labelled joints (sub-category: score = 3, k = 3).

(TIF)

Click here for additional data file.

Comparison between the calculated scores for unaffected and affected labelled joints using FOIAS (k = 5).

(A) Comparison between affected and unaffected labelled joints (sub-category: affected, k = 5). (B) Comparison between score = 1 and unaffected labelled joints (sub-category: score = 1, k = 5). (C) Comparison between score = 2 and unaffected labelled joints (sub-category: score = 2, k = 5). (D) Comparison between score = 3 and unaffected labelled joints (sub-category: score = 3, k = 5).

(TIF)

Click here for additional data file.

Comparison between the calculated scores for unaffected and affected labelled joints using FOIAS (k = 9).

(A) Comparison between affected and unaffected labelled joints (sub-category: affected, k = 9). (B) Comparison between score = 1 and unaffected labelled joints (sub-category: score = 1, k = 9). (C) Comparison between score = 2 and unaffected labelled joints (sub-category: score = 2, k = 9). (D) Comparison between score = 3 and unaffected labelled joints (sub-category: score = 3, k = 9).

(TIF)

Click here for additional data file.

15 Oct 2021

An objective, automated and robust scoring using fluorescence optical imaging to evaluate changes in micro-vascularisation indicating early arthritis

PLOS ONE

Dear Dr. Zerweck,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Please submit your revised manuscript by Nov 29 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at plosone@plos.org. When you're ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter.

If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at https://plos.org/protocols?utm_medium=editorial-email&utm_source=authorletters&utm_campaign=protocols.

We look forward to receiving your revised manuscript.

Kind regards,

Wei-Yen Hsu, Ph.D.

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. In your Methods, please provide a description of how patients were recruited to your study.

3. Please provide additional details regarding participant consent. In the ethics statement in the Methods and online submission information, please ensure that you have specified (1) whether consent was informed and (2) what type you obtained (for instance, written or verbal, and if verbal, how it was documented and witnessed). If your study included minors, state whether you obtained consent from parents or guardians. If the need for consent was waived by the ethics committee, please include this information.

If you are reporting a retrospective study of medical records or archived samples, please ensure that you have discussed whether all data were fully anonymized before you accessed them and/or whether the IRB or ethics committee waived the requirement for informed consent. If patients provided informed written consent to have data from their medical records used in research, please include this information.

4. Thank you for stating the following financial disclosure: "The authors declare that there was funding from the Fraunhofer Excellence Cluster for Immune mediated diseases  (https://www.cimd.fraunhofer.de/en.html) . The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript."

We note that one or more of the authors is affiliated with the funding organization, indicating the funder may have had some role in the design, data collection, analysis or preparation of your manuscript for publication; in other words, the funder played an indirect role through the participation of the co-authors. If the funding organization did not play a role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript and only provided financial support in the form of authors' salaries and/or research materials, please do the following:

a. Review your statements relating to the author contributions, and ensure you have specifically and accurately indicated the role(s) that these authors had in your study. These amendments should be made in the online form.

b. Confirm in your cover letter that you agree with the following statement, and we will change the online submission form on your behalf:

“The funder provided support in the form of salaries for authors [insert relevant initials], but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the ‘author contributions’ section.

5. We note that you have stated that you will provide repository information for your data at acceptance. Should your manuscript be accepted for publication, we will hold it until you provide the relevant accession numbers or DOIs necessary to access your data. If you wish to make changes to your Data Availability statement, please describe these changes in your cover letter and we will update your Data Availability statement to reflect the information you provide.

6. Please include your full ethics statement in the ‘Methods’ section of your manuscript file. In your statement, please include the full name of the IRB or ethics committee who approved or waived your study, as well as whether or not you obtained informed written or verbal consent. If consent was waived for your study, please include this information in your statement as well.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

Reviewer #3: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The Authors describe a method of assessing FOIAS interpretation automatically, with hand crafted features and clustering algorithms combined. They show that as a distribution of mild and critical patients and of healthy individuals are distinguishable by the method.

However, in order to prove that the method is able to provide some probability of joints belonging into one or the other diagnostic set, it is advisable that the Authors perform a patient by patient, and/or joint by joint analysis in the spirit of the followings:

Discriminating between distributions based on their median or some other statistical summary technique with confidence intervals is not sufficient for diagnosis. However it is sufficient to show, that there is an effect in the cohort that indicates that the features and analytics describe diagnostically valid distinctions.

Instead I suggest they do a machine learning classificator over the features, using the labeling provided in Fig 7. e.g. for classes. What is important, that there is a cross-validation scheme employed as well, prediction results on test data should be the main results. Classificators will yield probabilities for each patient and/or joint and/or pixels belonging to each category of disease or healthy.

Other than that the study is very interesting and important. It is likely that the future works mentioned by the authors and others are necessary to achieve reasonable single patient/joint/pixel classification, and I highly encourage the authors to do these extra analysis now.

Reviewer #2: The manuscript proposes a method for automatically and objectively analyze the measured fluorescence intensity generated by Fluorescence Optical Imaging to evaluate the amount of inflammation of each of the subject’s joints focusing on the distinction between normal joint status or arthritis in psoriatic arthritis patients compared to healthy volunteers.

The research results show that the proposed methodology can present a tendency towards a classification into unaffected and inflamed joints, although wide research is required to transform that tendency into a robust classification model.

I find the topic interesting and being worth of investigation and the document is well strucutred, organized, fluidly written, the methodology followed is clearly explained, the results are clearly presented and support the conclusions.

Although I propose the following suggestions / considerations:

- I strongly suggest authors from refraining using personal pronouns such as "we" and "our" throughout the text and I encourage them to write it in an impersonal form of writing.

- How were the 12 healthy volunteers selected, are they statistically representative to act as controls for the sample of 163 patients used.

- Would the results improve if some artificial intelligence techniques such as machine learning classifiers or deep learning models are used, this could be addressed at the discussion.

- Only 6 out of 16 references are of the last 5 years, more recent relevant references are advised to be included.

Reviewer #3: The author have implemented the novel work entitled An objective, automated and robust scoring using fluorescence optical imaging to evaluate changes in micro-vascularisation indicating early arthritis" in a systematic way.But there are few queries need to be addressed.

1. Quantitative results need to be included in the result section of abstract.

2. The aim and objectives should be refined and need to be included at the end of the introduction section.

3. It was mentioned that "After 10 s the prepared colour agent Indocyanine green (ICG) is 75

injected into the patient with a dose of 0.1 mgICG/kgbody weight". which area of interest the dye is injected? Imaging is taken after how many hours of dye injection?

4. Whether dye is injected for healthy controls?

5. How do you quantitatively evaluate the micro vascularization in arthritis ? Do you have any parameters to describe?

6. Whether the images for arthritis has well as healthy subjects were displayed?

7.What does figure 2d represents? It seems to be figure caption given in the article and figure2 doesn't match

8. How do you match the concentric circles into the finger joints as given in figure 5? because the size of the finger joints will differ for each patient?

9. Discussion section is completely missing. Need to be included

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: Yes: RICARDO VARDASCA

Reviewer #3: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at figures@plos.org. Please note that Supporting Information files do not need this step.

9 Feb 2022

Editor 1.)

Q: Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

A: The supporting information have been moved to the very end of the manuscript and are now following the references.

Editor 2.)

Q: In your Methods, please provide a description of how patients were recruited to your study.

A: Patients / volunteers were recruited from two prospective non-interventional studies (one with inclusion of patients diagnosed with psoriasis or psoriatic arthritis and one with inclusion of healthy volunteers without complaints in the joints of the hands).

This information was added to the subsection “Patient Group”.

Editor 3.)

Q: Please provide additional details regarding participant consent. In the ethics statement in the Methods and online submission information, please ensure that you have specified (1) whether consent was informed and (2) what type you obtained (for instance, written or verbal, and if verbal, how it was documented and witnessed). If your study included minors, state whether you obtained consent from parents or guardians. If the need for consent was waived by the ethics committee, please include this information.

If you are reporting a retrospective study of medical records or archived samples, please ensure that you have discussed whether all data were fully anonymized before you accessed them and/or whether the IRB or ethics committee waived the requirement for informed consent. If patients provided informed written consent to have data from their medical records used in research, please include this information.

A: All patients / volunteers provided signed informed consent for inclusion and were fully capable to give informed consent for participation in the study.

This information was added to the subsection “Patient Group”.

Editor 4.)

Q: Thank you for stating the following financial disclosure: "The authors declare that there was funding from the Fraunhofer Excellence Cluster for Immune mediated diseases (https://www.cimd.fraunhofer.de/en.html) . The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript."

We note that one or more of the authors is affiliated with the funding organization, indicating the funder may have had some role in the design, data collection, analysis or preparation of your manuscript for publication; in other words, the funder played an indirect role through the participation of the co-authors. If the funding organization did not play a role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript and only provided financial support in the form of authors' salaries and/or research materials, please do the following:

a. Review your statements relating to the author contributions, and ensure you have specifically and accurately indicated the role(s) that these authors had in your study. These amendments should be made in the online form.

b. Confirm in your cover letter that you agree with the following statement, and we will change the online submission form on your behalf:

“The funder provided support in the form of salaries for authors [insert relevant initials], but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. The specific roles of these authors are articulated in the ‘author contributions’ section.

A: Option B was chosen and the statement was added to the cover letter.

Editor 5.)

Q: We note that you have stated that you will provide repository information for your data at acceptance. Should your manuscript be accepted for publication, we will hold it until you provide the relevant accession numbers or DOIs necessary to access your data. If you wish to make changes to your Data Availability statement, please describe these changes in your cover letter and we will update your Data Availability statement to reflect the information you provide.

A: The data was uploaded and is accessible. The DOI is:

10.5281/zenodo.5705208

And the target url:

https://doi.org/10.5281/zenodo.5705208

Editor 6.)

Q: Please include your full ethics statement in the ‘Methods’ section of your manuscript file. In your statement, please include the full name of the IRB or ethics committee who approved or waived your study, as well as whether or not you obtained informed written or verbal consent. If consent was waived for your study, please include this information in your statement as well.

A: Ethical approval was received from the ethics committee of the University Hospital Frankfurt a. Main, Germany. All patients / volunteers provided signed informed consent.

This information was added to the subsection “Patient Group”.

Reviewer #1 1.)

Q: I suggest they do a machine learning classificator over the features, using the labeling provided in Fig 7. e.g. for classes. What is important, that there is a cross-validation scheme employed as well, prediction results on test data should be the main results. Classificators will yield probabilities for each patient and/or joint and/or pixels belonging to each category of disease or healthy.

A: As suggested a machine learning classifier (random forest) for the clinical label as well as the FOIAS label has been trained and tested.

For both cases classes are not balanced. Therefore, all class sizes have been cropped to the smallest class size (for example for the FOIAS classifier the number of score 3 joints is 18). For each run samples from the classes containing more than the minimum amount of examples have been picked randomly. Due to the cropping many samples remain unconsidered for each run. Thus, the classifiers have been trained and tested 500 times.

Since for the clinical labeling a joint can be labeled as healthy, swollen, tender or swollen and tender the classifier has been trained to distinguish only between unaffected or affected scores. Thus, a binary classification problem has to be solved. The calculated F1 scores range from around 40 % to around 62 % detection rate.

For the FOIAS labeling a 4 class (Score 0, 1, 2 or 3) classification problem has to be solved. The calculated F1 scores range from 0 % to 66 % detection rate.

These fluctuating results are based on the overlap of scores for unaffected and affected joints and motivate to understand the causes of the heterogeneity of the data in more detail, since a classification with the current data seems extremely difficult. Therefore, we are currently conducting a study investigating the non-disease related impact factors on the signal (for example temperature, BMI, alcohol consumption and many more).

A subsection describing the machine learning approach has been added as well as a paragraph in the discussion describing the results.

Reviewer #2 1.)

Q: I strongly suggest authors from refraining using personal pronouns such as "we" and "our" throughout the text and I encourage them to write it in an impersonal form of writing.

A: The relevant passages have been changed.

Reviewer #2 2.)

Q: How were the 12 healthy volunteers selected, are they statistically representative to act as controls for the sample of 163 patients used.

A: The 12 healthy volunteers were selected with regards to the following in- and exclusion criteria: (a) missing musculoskeletal complaints on the hands, (b) missing diagnosis of joint diseases, (c) missing comorbidities with focus on diseases that go ahead with joint diseases, (d) missing contraindications to use indocyanine green as colour agent for FOI examinations. Due to these criteria, the 12 volunteers are not affected by PsA and can act as a control group.

This passage has been added to the manuscript.

Reviewer #2 3.)

Q: Would the results improve if some artificial intelligence techniques such as machine learning classifiers or deep learning models are used, this could be addressed at the discussion.

A: Refer to reviewer 1.

Reviewer #2 4.)

Q: Only 6 out of 16 references are of the last 5 years, more recent relevant references are advised to be included.

A: We would have liked to include more recent works. But, to our knowledge there have not been many approaches to automatically evaluate the FOI images in the recent years. Additionally, in this work we have not been using machine learning approaches in the computer vision domain. In our current research, we shift our focus to a more machine learning based approach (especially for the segmentation task), which will also lead to more recent references.

Reviewer #3 1.)

Q: Quantitative results need to be included in the result section of abstract.

A: In this work a method to possibly overcome the heterogeneity in the data is presented and a tendency towards this goal was achieved. This result is not expressed in numbers.

Reviewer #3 2.)

Q: The aim and objectives should be refined and need to be included at the end of the introduction section.

A: The last paragraph of the introduction has been rephrased to emphasis the aim and objectives of the study.

Reviewer #3 3.)

Q: It was mentioned that "After 10 s the prepared colour agent Indocyanine green (ICG) is 75 injected into the patient with a dose of 0.1 mgICG/kgbody weight". which area of interest the dye is injected? Imaging is taken after how many hours of dye injection?

A: The dye is injected into the arm (it was added in the manuscript).

The dye is injected 10 s after the Xiralite machine starts taking the images. Thus, the images are taken while adding the dye.

Reviewer #3 4.)

Q: Whether dye is injected for healthy controls?

A: Yes, the dye is the fluorescent substance and thus, essential for each measurement. For clarification, the word “patient” was changed to the word “proband” in the “Data acquisition” section and one more sentence mentioning this was added.

Reviewer #3 5.)

Q: How do you quantitatively evaluate the micro vascularization in arthritis ? Do you have any parameters to describe?

A: Clinical examination can be used to detect signs of changes in vascularization in arthritis that go along with specific signs of inflammation (including tumor, calor, dolor, rubor, function laesa). A reliable method to measure changes in micro-vascularisation that is not invasive such as angiography which is limited by x-ray load is not available in clinical routine care. So, we used clinical examination (joint assessment) as control assessment for the comparison between the groups.

Reviewer #3 6.)

Q: Whether the images for arthritis has well as healthy subjects were displayed?

A: Throughout, the presented work all images (Figure 1, 2, 4 and 5) refer to the same exemplary patient. This patient is not a healthy volunteer. Adding images of a healthy volunteer would not add any additional information.

Reviewer #3 7.)

Q: What does figure 2d represents? It seems to be figure caption given in the article and figure2 doesn't match

A: Figure 2d represents the extracted joint areas as a binary image. Figure 5 visualizes an overlay of the heatmap with these areas.

Reviewer #3 8.)

Q: How do you match the concentric circles into the finger joints as given in figure 5? because the size of the finger joints will differ for each patient?

A: Three different cases have to be differentiated:

1.) Joints located within in the fingers (DIP, PIP, IP):

The radius is defined as the smallest distance to the edge of the binary hand mask. Thus, the radius differs due to the width of the finger at the given location.

2.) Joints located within the hand (MCP), but omitting the Thumb’s MCP joint:

The radius is defined as the smallest distance to either the edge of the binary hand mask or the distance to the nearest other MCP location. Thus, these four MCP circles are matched to the hand’s width and to the individual hand

3.) Thumb’s MCP joint:

The size of this joint is equal to the size of the Thumb’s IP joint.

Reviewer #3 9.)

Q: Discussion section is completely missing. Need to be included

A: The results are discussed in the section “Evaluation and conclusion”. However, the chapter’s name was renamed for a clearer distinction.

Submitted filename: Response to Reviewers.docx

Click here for additional data file.

1 Sep 2022

An objective, automated and robust scoring using fluorescence optical imaging to evaluate changes in micro-vascularisation indicating early arthritis

PONE-D-21-05449R1

Dear Dr. Zerweck,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Diego Raimondo

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #2: All comments have been addressed

Reviewer #3: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #2: Yes

Reviewer #3: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #2: Yes

Reviewer #3: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #2: Yes

Reviewer #3: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #2: Yes

Reviewer #3: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #2: The manuscript aims to automatically and objectively analyze the measured fluorescence intensity generated by Fluorescence optical imaging technique to evaluate the amount of inflammation of each of the subject’s joints focusing on the distinction between normal joint status or arthritis in psoriatic arthritis patients compared to healthy volunteers.

With the proposed method a tendency towards a classification into unaffected and inflamed joints can be achieved, but further research is necessary to transform the tendency into a robust classification model.

I find the topic interesting and being worth of investigation and the document is well strucutred, organized, fluidly written, the background is adequate, the methodology well explained (formulas are correct), results are clearly presented, supporting the discussion and conclusions.

I am happy with the authors' answers and action towards the reviewers questions and comments and significantly have improved the manuscipt which I support its acceptance for publication at Thermal Biology journal.

Reviewer #3: The authors addressed the queries raised by the reviewer. Hence the article can be accepted in its current form.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #2: No

Reviewer #3: No

**********

6 Sep 2022

PONE-D-21-05449R1

An objective, automated and robust scoring using fluorescence optical imaging to evaluate changes in micro-vascularisation indicating early arthritis

Dear Dr. Zerweck:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Diego Raimondo

Academic Editor

PLOS ONE