Prediction of the Activity of Early Ankylosing Spondylitis using Radiomics Texture Analysis on STIR

doi:10.21203/rs.3.rs-2572727/v1

Purpose

To evaluate the early activity of ankylosing spondylitis(AS)is very important for clinical decision. Owing to lacking of uniform standard and some other limitations, the methods for it are still difficult to practice by clinicians. The study aimed to explore the value of texture analysis of radiomics based on STIR to evaluate the activity of bone marrow edema of sacroiliac joint in early AS.

Methods

A total of 43 patients with early AS were collected in our hospital, including 30 males and 13 females. According to the ratio of 7:3, all the patients were randomly divided into the training cohort (n = 116) and verification cohort (n = 56). Optimal feature subsets were obtained by the Mann-Whitney U test, the minimum-Redundancy Maximum-Relevancy (mRMR), and then the least absolute shrinkage and selection operator (LASSO) using these texture feature parameters. The optimal feature subset was used to construct the final prediction model, and Radscore was obtained by summing the selected features after being weighted by their coefficients. The effectiveness of Radscore value, SPARCC score and ADC value was evaluated by ROC curve. The correlations between Radscore values and the values of SPARCC score, ADC, BASDAI, ESR, CRP, ASDAS-ESR, ASDAS-CRP were analyzed by the Spearman correlation test.

Results

The eight texture feature subsets were selected to obtain Redscore. In the training cohort, to differentiate early AS sacroiliac joint bone marrow edema between between active and stable group,the AUC values of Radscore, SPARCC score and ADC were 0.81,0.91,0.78, respectively. In the validation cohort, the AUC values of them were 0.87,0.89,0.85. In the two cohorts, there were no significant differences in AUC values between values of Radscore and SPARCC score, ADC (P > 0.05). There was a significant difference in AUC values betweenvalues of SPARCC score and ADC in the training cohort (P < 0.05), without statistical significance in the validation cohort (P > 0.05). The correlations were all low between Radscore values and the values of the ESR, CRP, BASDAI, ASDAS-ESR, ASDAS-CRP (P < 0.05).

Conclusion

Radiomics analysis based on STIR texture analysis has a good prediction for the evaluation of bone marrow edema activity of sacroiliac joint in AS.It can be a new noninvasive and objective evaluation method for AS activity, especially for the invisible lesions.

STIR

Radiomics

texture analysis

ankylosing spondylitis

Activity

Evaluation of the activity of AS is key in guiding clinical management and intensive therapy by drug combination. At present, there is no gold standard evaluating the activity of AS, and several common evaluation indexes, including BASDAI, ESR, CRP, ASDAS-ESR, ASDAS-CRP, all have their own limitations. BASDAI can be influenced by many factors of patients, such as the cultural and educational background, level of understanding, judgment and pain threshold, and bias for BASDAI score caused by judgement of different doctors. ASDAS-CRP and ASDAS-ESR, proposed by Assess Spondyloarthritis International Society (ASAS), were two new comprehensive assessment methods.However, owing to complicated calculation formula, these activity indexes are not received widely by clinicians[1, 2, 3, 4]. In addition, some imaging examinations, such as X-ray, CT, US and MRI, can benefit a lot for the diagnosis of AS, though each with its own limitations. X-ray and CT could clearly show the changes of bone structure, but could not be well used to diagnose early AS and evaluate the activity of AS [5, 6]. Owing to the overlap of bones, Us usually fail to fully display the anatomical structures and subtle changes of sacroiliac joint[7]. Widely recognized as the most sensitive and specific imaging tool for the diagnosis of AS, MRI is excellent to evaluate AS activity, with its common methods of SPARCC score, DWI, IVIM and DCE-MRI sequences[8]. SPARCC score, based on the observation of bone marrow edema on STIR sequence, has been widely studied and clinically recognized; However, many factors depending on operators may affect its accuracy. As a rapidly advanced image analysis technique, radiomics has attracted increased attention in recent years. Radiomics can extract high throughput features of the segmented regions and be further selected to form a model to quantitatively analyse the lesion heterogeneity[9]. Radiomics has been gradually studied and applied in the osteomuscular system to observe the microstructures of bones[10]. However, whether radiomics analysis based on STIR imaging could evaluate the activity of AS is still unclear. Therefore, this study aimed to investigate the value of radiomics-clinical analysis based on STIR features in the evaluation of bone marrow edema activity of sacroiliac joint in AS.

Patients

This retrospective study was approved by the institutional review board and informed consent was signed. A total of 43 patients were retrospectively analysed with early AS confirmed by Rheumatology and Immunology Department in our hospital, including 30 males and 13 females, aged from 14 to 51 years, with a median age of (27.52 ± 4.28) years. The inclusion criteria were (1) these patients met the diagnostic criteria of spinal arthropathy revised by Assessment of spondyloarthritis interationnal society (ASAS) experts in 2009[11, 12]; (2) they must met the early diagnostic criteria of AS: X-ray or CT examination showed slight or no obvious bone changes in sacroiliac joint bones without stenosis of joint spaces; the grade was below II on MRI; (3) they should have clinical related activity evaluation indexes: BASDAI, HLA-B27, ESR and CRP; (4) no anti-rheumatic immune system related drugs were taken orally within 3 months; (5) there were no absolute contraindications of MRI. The exclusion criteria were (1) patients with AS combined with rheumatic immune system diseases, tumors, traumas or infections; (2) various artifacts leading to poor image quality made it unable to accurately outline the lesions.

All study participants underwent MRI of the spine with a 3.0T system(Discovery MR750,GE Healthcare, Milwaukee, WI, USA).The patient's belongings were removed before scanning, such as mobile phone, key, watch, magnetic card, etc. Parameters of Axial FSE T1-weighted imaging were as follows: repetition time (TR) = 400 ms; echo time (TE) = 10ms; slice thickness = 4 mm; spacing = 1mm; field of vision(FOV )280mm×280mm. Axial FSE T2-weighted imaging were as follows: TR = 2400ms; TE = 110ms; slicethickness = 4mm; spacing = 1mm;FOV = 280mm×280mm. Oblique coronal short tau inversion recovery(STIR): TR = 2700ms, TE = 110ms, slice thickness = 4mm, spacing = 1.0mm, FOV = 280mm×280mm. Axial diffusion weighted imaging(DWI): TR = 5000ms,TE = 50ms, layer thickness = 4mm, spacing = 1.5mm, FOV = 380mm×380mm, number of excitation(NEX) = 4, b values = 0,600s/mm. All MRI examinations were processed by two experienced radiologist in musculoskeletal imaging, corrected by eddy current or motion correction. In cases with disagreement, scans were reviewed by both readers to rejudge the result.

Specific Criteria For Clinical Evaluation Of Sacroiliac Arthritis

The evaluation criteria of BASDAI (Bath ankylosing spndylitis disease activity index) scores were as follows[13]: ① pain or discomfort level in the neck, back or hip joint; ② degree of swelling, pain or discomfort in joints other than the neck, back or hip joint; ③ degree of tenderness in different parts; ④ degree of general tiredness and discomfort; ⑤ degree of stiffness in the lumbar and back in the morning; as well as ⑥ duration of back stiffness after rising in the morning. Among them, criteria of ①-⑤ were designed as 0–10 points according to the degree of pain or discomfort, and "no any pain sensation" was 0 score, "unable to bear pain with imagination" was 10 scores. However, for ⑥, the results was set as 0 h (hour) = 0 point, 0.5 h = 2 points, 1 h = 4 points, 1.5 h = 6 points, 2 h = 8 points, and 2.5 h = 10 points, and was defined as the score of morning stiffness= (the degree of morning stiffness + duration of morning stiffness)/2. The ①-⑥ scores were set as A-F, with 10 points for each item up to a total of 60 points. BASDAI scoring was set as 0–10 points, using the formula: BASDAI scoring = 0.2×[A + B + C + D + 0.5× (E + F)]. Based on the above criteria, the BASDAI score of ≥ 6 points indicated active stage, while the score ≤ 4 points indicated stable stage. Patients with a score between 4 and 6 points need to be evaluated using serological indexes. The values of ESR ≥ 30 mm/h (Westergren) or CRP ≥ 3 mg/L were showed in the active stage. According to the BASDAI score, 43 patients were divided into active group (28 cases) and stable group (15 cases).

HLA-B27 was positive (+) and negative (-) by flow cytometry. The normal reference values of CRP equipment adopts bc-5390 CRP model were 0-3mg/L. The normal reference values of the ESR adopts ALIFAX TEST₁ equipment were 0–15 mm/h for men and 0–20 mm/h for women. ASDAS-ESR and ASDAS-CRP were computed as follows[14]:ASDAS-ESR = 0.079×(pain of lumbar and back) + 0.069×(duration of morning stiffness) + 0.113×(overall evaluation of patients) + 0.086×(peripheral arthralgia) + 0.293×√ESR;while ASDAS-CRP = 0.121×(pain of lumbar and back) + 0.058× (duration of morning stiffness) + 0.11×(Overall evaluation of patients) + 0.073×(peripheral pain relief/swelling) + 0.579×Ln (CRP + 1). The overall evaluation of the patients was based on the visual analogue scale (VAS) with a length of 10cm. We analysed the degree using a scale of 10 points, with 10 points referring to the most intolerable pain or discomfort. The above indexes were independently completed by two experienced rheumatologists, and the average value was taken as the final clinical evaluation index.

Mri Evaluation Of Sacroiliac Arthritis

In this paper we mainly discussed the SPARCC score and the ADC value to evaluate the activity of sacroiliac arthritis. The SPARCC scoring system required oblique coronal T2W STIR imaging of the sacroiliac joints[15]. Each sacroiliac joint was separated to 4 quadrants ,including upper/ lower iliac and upper /lower sacral quadrants, and was scored in 6 consecutive slices, in which 1 score denoted an increased signal and 0 score meant a normal signal, with a maximum score of 48; In addition, the depth (≥1 cm = 1 point)and signal intensity (≥ signal in presacral blood vessels = 1 point) of bone edema were separately scored for each slice, with maximum 12 points. SPARCC score was the sum of the range, signal intensity and depth of marrow edema, with a maximum of 72 points.

Combined with STIR sequence, the marrow edema sections in the sacroiliac joint were selected as ROI on DWI. we used Ge MRI advantage workstation 4.6 to process the ADC measurement data. We ensured the measurement was consistent with the lesion scope, and avoided joint space, osteosclerotic, cystic necrosis and fat areas. The ADC values for right ilium, right sacrum, left ilium or left sacrum were independently measured, and then the average values for each side were calculated. Absence of the marrow edema prompted selection of the best three consecutive layers of the sacroiliac joint. In addition, 12 bone marrow sections under iliac and sacroiliac joint surfaces were measured and then the average was computed and used as the final value. The measurement of the above values was completed by two experienced MRI diagnostic physicians, and the average value of the measurement results was taken as the evaluation index of MRI.

Roi Segmentation

All patients were divided into active group (28 cases) and stable group (15cases) according to BASDAI activity evaluation criteria. The STIR images of all patients were imported into ITK-snap software in DICOM format, and the ROI of the lesions was outlined (see Fig. 1). We outlined the iliac and sacral sides of the left and right sacroiliac joints separately, including four ROI areas (right ilium, right sacrum, left ilium and left sacrum), and 172 ROI regions were outlined in 43 patients.

Radiomic Features Extraction And Establishment Of Omics Model

The original STIR and ROI images were imported into software of GE.A.K. and texture feature parameters were automatically extracted by this software. According to the ratio of 7:3, all the patients were randomly divided into training cohort (n = 116) and verification cohort (n = 56). The training cohort was mainly used for texture feature selection and model building, and the verification cohort was mainly used for model effectiveness evaluation. Optimal feature subsets were obtained by the Mann-Whitney U test, the minimum-Redundancy Maximum-Relevancy (mRMR), and then the least absolute shrinkage and selection operator (LASSO) using these texture feature parameters (see Fig. 2). The optimal feature subset was used to construct the final prediction model, and Radscore was obtained by summing the selected features after being weighted by their coefficients.

A: finding the optimal hyperparameter λ value through 10-fold cross-validation to minimize the deviation of the model (the left vertical dotted line represents the Log(λ)value corresponding to the best λ value).

B: a relation diagram depicting the changes of different texture features as the λ value changed (the dotted line corresponded to the broken line on the left of figure A); under the condition of this λ value, as coefficient were not 0, the retained features were the ones screened by LASSO.

Statistical analysis

The count data were analysed by χ² test. All Measurement data were expressed using median ± quartile interval and detected by the Mann-Whitney U test. The effectiveness of Radscore value, SPARCCscore and ADC value was evaluated by ROC curve. The difference of area under ROC curve of three methods was compared by the Delong's test. The correlations between Radscore values and the values of SPARCC, ADC, BASDAI, ESR, CRP, ASDAS-ESR ASDAS-CRP were analysed by the Spearman correlation test.

Clinical characteristics

For 43 patients, including 28 cases in active group and 15 cases in stable group ,we outlined the 172 ROI of lesions in bilateral sacroiliac joints, including 80 lesions in the iliac side, 32 lesions in the sacral side in the active group and 36 lesions in the iliac side, 32 lesions in the sacral side in the stable group. All ROI features were randomly assigned to the training cohort (n=116) and the validation cohort (n=56) in the ratio of 7:3. Table 1 showed the clinical characteristics of patients in the two cohorts. There were no significant differences in age and gender between the two cohorts. ESR values were no significant differences in the validation cohort. However, the values of HLA-B27, ESR, CRP, ASDAS-ESR、ASDAS-CRP、SPARCC scores and ADC were statistically significant in the two cohorts.

Table 1 Clinical features of patients in the training and validation cohorts

Feature	Cohort		Active group (n=80/32）	Stable group （n=36/24）	Z（χ2）	P
Age （year）		Training	29.00±13.25	27.00±9.00	-0.144	0.886
Age （year）		Validation	24.5±20.75	31.00±4.00	-2.265	0.240
Gender （Male /Female）		Training	56/24	24/12	1.29^a	0.720
Gender （Male /Female）		Validation	28/4	16/8	3.535^a	0.060
HLA-B27 （+/-）		Training	72/8	36/0	3.867^a	0.046
HLA-B27 （+/-）		Validation	32/0	20/4	5.744^a	0.017
ESR （mm/h）		Training	22.5±15.5	9.00±9.00	-5.115	0.000
ESR （mm/h）		Validation	19.00±19.50	16.00±7.00	-1.599	0.110
CRP （mg/L）		Training	12.05±16.96	2.12±1.11	-7.166	0.000
CRP （mg/L）		Validation	12.77±18.38	3.5±2.36	-5.842	0.000
ASDAS-ESR		Training	2.87±1.28	1.64±0.63	-6.019	0.000
ASDAS-ESR		Validation	2.46±1.32	1.82±0.33	-5.576	0.000
ASDAS-CRP		Training	3.48±1.80	1.51±0.54	-7.070	0.000
ASDAS-CRP		Validation	2.85±1.49	1.37±1.11	-6.107	0.000
BASDA （score）		Training	5.79±1.28	4.70±0.70	-5.788	0.000
BASDA （score）		Validation	5.55±0.73	5.00±0.80	-5.068	0.000
SPARCC （score）		Training	37.5±20.00	17.00±3.00	-7.077	0.000
SPARCC （score）		Validation	33.00±23.25	15.5±8.00	-4.918	0.000
ADC（×10^-4mm²/s）		Training	6.43±4.00	5.12±1.93	-4.810	0.000
ADC（×10^-4mm²/s）		Validation	7.02±4.27	4.61±1.41	-4.470	0.000

The Mann-Whitney U test and chi-squared test for comparison of all values. HLA-B27: Human Leukocyte Antigen-B27;ESR: Erythrocyte Sedimentiation Rate; CRP: C-reactive Protein; ASDAS: Ankylosing Spondylitis Disease Activity Score; BASDAI: Bath Ankylosing Spondylitis Disease Activity Score; SPARCC:Spondyloarthritis Research Consortium of Canada;ADC: Apparent Diffusion Coefficient;

Radscore building and diagnostic validation based on STIR

Eight texture features of 895 radiomics features subsets were selected to establish the Rad-score with the Mann Whitney U test, mRMR and LASSO regression model (Fig3).

Radscore=0.823*STIR_wavelet_HHH_gldm_SmallDependenceLowGrayLevel+0.456*STIR_original_gldm_Large Dependence High Gray Level+0.396*STIR

_wavelet_LHH_glrlm_Gray Level Non Uniformity+0.321*STIR_wavelet_LLH

_glcm_Correlation+0.067*STIR_wavelet_LLL_glszm_Large Area High Gray Level+ -0.089*STIR_wavelet_HHH_ngtdm_Strength+-0.279*STIR_wavelet_LLH_glcm_ Maximal Correlation Coefficient+-0.382*STIR_wavelet_ HHH_ glcm_ Cluster Shade

ROC curve analysis of Radscore, SPARCC score and ADC value

In the training cohort, the optimal cut-off values (AUC, AC,SEN,SPE,PPV,NPV) for differentiating early as sacroiliac joint bone marrow edema between active group and stable group were Radscore=51.46(0.81,76.72%,81.25%,66.67%,84.42%,61.54%), SPARCC =20scores(0.91,89.66%,90.00%,88.89%,94.97%,80.00%),

ADC=6.71×10^-4mm²/s（0.78,2.93%,48.75%,94.44%,95.12%,45.33%).

In the validation cohort, the optimal cut-off values (AUC,AC,SEN,SPE,PPV,NPV) for differentiating early as sacroiliac joint bone marrow edema between active and stable group were Radscore=51.46(0.87,82.14%,90.65%,70.83%,80.56%,85.00%),

SPARCC=20score(0.89,78.57%,87.50%,66.67%,77.78%,80.00%),

ADC=6.71×10^-4mm²/s（0.85,71.42%,53.13%,95.83%,94.44%,60.53%).

In the training and validation cohorts, there were no significant differences in AUC between Radscore and SPARCC score, ADC value (P>0.05). There were significant differences in AUC in the training cohort between SPARCC score and ADC value (P<0.05), without statistical significance in the validation cohort (P>0.05). (Table 2,3, Fig 4)

Table 2 Diagnostic efficiency of different models in the training and validation cohorts

Model	cohort	TV	AUC	AC (%)	SEN (%)	SPE (%)	PPV (%)	NPV (%)
Radscore	training	51.46	0.81	76.72	81.25	66.67	84.42	61.54
Radscore	validation	51.46	0.87	82.14	90.63	70.83	80.56	85.00
SPARCC	training	20.00	0.91	89.66	90.00	88.89	94.74	80.00
SPARCC	validation	20.00	0.89	78.57	87.50	66.67	77.78	80.00
ADC	training	6.71	0.78	62.93	48.75	94.4Accuracy4	95.12	45.33
ADC	validation	6.71	0.85V5	71.42	53.13	95.83	94.44	60.53

The effectiveness of three methods were evaluated by ROC curve. TV: Threshold Value, AUC: Area Under Curve, AC: Accuracy; SEN: Sensitivity,SPE: Specificity, PPV: Positive Predictive Value, NPV: Negative Predictive Value

Table 3 ROC curve comparisons among different models (Delong's test)

Model	Training cohort		Validation cohort
Model	Z P		Z P
Radscore-SPARCC	1.867	0.062	0.250	0.802
Radscore-ADC	0.499	0.618	0.311	0.756
SPARCC-ADC	3.591	0.000	0.770	0.441

ROC curves were compared among different models by Delong's test ,which were not statistically significant（p＞0.05）.

Correlations of between Radscore，MRI and clinical activity indexes

There were low correlations between Radscore values and SPARCC scores, ADC values, and the correlation coefficients were both 0.29 (P<0.05).

There were all low correlation between Radscore values and the values of the ESR, CRP, BASDAI, ASDAS-ESR, ASDAS-CRP, and the correlation coefficients were 0.28, 0.21, 0.37, 0.28 and 0.36, respectively (P<0.05). (Table 4)

Table4 Correlations between Radscore and clinical indicators (Spearman's rho)

Radscore

BASDAI

ESR

CRP

ASDAS

-ESR

ASDAS-CRP

SPARCC

ADC

r

0.28

0.21

0.37

0.28

0.36

0.29

P

0.000

0.005

4e^-07

0.000

1.1e^-06

0.000

The correlations of all values were analysed by the Spearman correlation test. Radscore :Radiomic Score; BASDAI: Bath Ankylosing Spondylitis Disease Activity Score;ESR: Erythrocyte Sedimentiation Rate; CRP:C-reactive Protein; ASDAS: Ankylosing Spondylitis Disease Activity Score; SPARCC:Spondyloarthritis Research Consortium of Canada;ADC: Apparent Diffusion Coefficient

The present study aimed to assess the value of radiomics-clinical analysis based on 3.0T STIR imaging in distinguishing the activity of bone marrow edema of sacroiliac joint in AS. Firstly, the study evaluated the effects of traditional methods of MRI (SPARCC score and ADC value) and clinical methods (BASDAI, ESR, CRP, ASDAS-ESR and ASDAS-CRP) for the activity of AS. Then, Rad-score was obtained on STIR imaging to evaluate the activity of bone marrow edema of sacroiliac joints. Finally, the differences and correlations of these three evaluation methods were analysed by the Delong's test and Spearman correlation test. Our results showed that the radiomics based on STIR imaging can be used to evaluate the activity of AS, with the highest AUC value of 0.81 in the training cohort and 0.87 in the validation cohort, showing excellent ability. However, there were significant differences for AUC with low correlations among the three methods.

STIR imaging can significantly inhibit the fat signal in bone marrow areas and more clearly show the signal of bone marrow edema, It has certain advantages for the evaluation of diagnosis, treatment and activity of AS [16]. Bone marrow edema is the earliest and most common pathological change of AS[17, 18], and is also an important sign to confirm this disease [16]. Landew é et al [19] proposed that SPARCC score had a high sensitivity and good specificity for the changes of AS in sacroiliac joints, so it can be used as the preferred method. In this study, we found that SPARCC score had statistical significance in the evaluation of AS activity with the highest AUC of 0.91 and 0.89 in training and validation cohorts. DWI, as a conventional MRI sequence, can reflect the pathological and physiological changes in living tissue[20]. ADC values can objectively quantify early bone marrow edema with the advantages of short-time scanning and no contrast agent[21]. In this study, we found that ADC values had statistical significance in the evaluation of AS activity with the highest AUC of 0.78 and 0.85 in training and validation cohorts. Therefore, it can be concluded that SPARCC score and ADC value have high reliability in evaluating the activity of AS using traditional methods on MRI.

In recent years, radiomics has been highlighted in clinical study and gradually applied to the study field of the osteomuscular joint system[22, 23, 24]. Fengxian, Wang et al.[25]compared the results of ADC histogram analysis and SPARCC score in sacroiliac joints between 57 AS patients and 27 healthy volunteers. Their study showed that ADC histogram parameters were better than SPARCC score in evaluating AS activity, with a good correlation between the two methods. It is widely recognized that bone marrow edema is the earliest and most common pathological changes of AS, and STIR imaging can more clearly show the signal of bone marrow edema because it can significantly inhibit the fat signal. Meanwhile, the SPARCC scoring system is operated based on the STIR sequence of sacroiliac joint, with high reliability. Therefore, we verified the value of SPARCC scoring by STIR sequence for the activity of AS bone marrow edema, and explored the value of texture analysis of radiomics based on STIR, two methods proving the feasibility of the STIR radiomics features for distinguishing the activity of bone marrow edema of sacroiliac joint in AS. We had extracted eight kinds of quantitative radiomics features of lesions based on the STIR, including the correlation, maximum correlation coefficient and cluster shadow of gray level co-occurrence matrix (GLCM), the large dependence high gray level and small dependence high gray level of gray-level dependence matrix features (GLDM), gray-Level Non-Uniformity and large-scale high gray of gray level run length matrix (GLRLM), and the strength feature in the Neighboring gray tone difference matrix (NGTDM). Finally, the eight features were selected to achieve Rad-score with a least absolute shrinkage and selection operator (LASSO).

In this study, we compared the effects of AUC of SPARCC score, ADC value and Rad-score in evaluating the activity of marrow edema in sacroiliac joints. We found SPARCC score had the highest AUC efficiency, followed by Radscore value and then ADC value. Through statistical comparisons of the three evaluation methods, there were no statistical differences in the validation cohort, indicating that the radiomics model based on STIR images had the similar value compared SPARCC score with ADC value in evaluating the activity of AS. In addition, the correlation analysis of the radiomics on STIR was made by SPARCC score and ADC value, with a low positive correlation. The reason may be that texture analysis of radiomics was a sign of the heterogeneity of tissue or lesions[26], while SPARCC score was to evaluate the signal level and range of bone marrow edema, and the ADC value reflected the diffusion and movement ability of water molecules in the bone marrow edema area on DWI; although these three methods have clear value in the evaluation of AS activity, the specific methods and the principles to reflect lesions were different. In conclusion, radiomics can provide quantitative information which can not be achieved by visual analysis and can be a new evaluation method for the activity of AS for clinic.

However, there were several limitations for the study. Firstly, this study was a single-center retrospective study with a small sample size and individual differences in patients' gender, age and disease conditions, so a larger sample with multicenter external validation should be needed in the next step. Secondly, the patients had no results of histopathological examinations,so it was impossible to comparatively analyse the changes between texture parameters and pathology in detail.

MRI	Magnetic resonance imaging
STIR	Short tau inversion recovery
DWI	Diffusion weighted imaging
ADC	Apparent diffusion coefficient
ROI	Region of interest
AS	Ankylosing spondylitis
SIJ	Sacroiliac joint
SPARCC	Spondyloarthritis research consortium of canada
BASDAI	Bath ankylosing spndylitis disease activity score index
HLA-B27	Human leukocyte antigen-B27
CRP	Creactive protein
ESR	Erythrocyte sedimentiation rate
ASAS	Assessment of spondyloarthritis interationnal society
ASDAS-CRP	Ankylosing spondylitis disease activity score-CRP
ASDAS-ESR	Ankylosing spondylitis disease activity score-ESR
LASSO	Least absolute shrinkage and selection operator
mRMR	minimum-Redundancy Maximum-Relevancy

Ethics Approval and Consent to Participate

The institutional review board of The Second Affiliated Hospital of Shandong First Medical University approved this retrospective study. In addition ，the need for obtaining signed informed consent was exempted by the ethics committee of The Second Affiliated Hospital of Shandong First Medical University because of retrospective analysis.All methods were carried out in accordance with relevant guidelines and regulations.

Consent for publish

Not applicable.

Availability of data and materials

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests

The authors declare that they have no competing interests.

Funding

Academic promotion programme of Shandong First Medical University (No. 2019QL017).

Comparative study of MR IDEAL-IQ and QCT in the evaluation of ankylosing spondylitis complicated with osteoporosis,( No.2020NS122)

Study on the changes of bone microenvironment in ankylosing spondylitis osteoporosis based on Multimodal fusion imaging technology(No. 202209010762)

Quantitative diagnostic value of IVIM-DWI and DCE-MRI in pathological changes of osteoporosis animal model(No.2020NS246)

Authors' Contributions

ZJS, ZY, MW, DTS, JC, HL,HZ,JKH,CQLand JQdesigned the study. JQ is the conceptualization, Rev HL iewing and Editing. ZJS is the fist auther, Writing-Original draft preparation. ZY help to modify English language,MW and JC performed the data analysis. HL is responsible for statistical analysis .HZ contributed to clinical data collection. DTS performed MRI operation. JKH and CQL supervised the project ,and added some modifications to the manuscript. All authors made contributions to its final form,and approved the manuscript in this form.

Acknowledgements

We apologize to those authors we have not been able to cite due to space constraints.

Authors’ Information

^aDepartment of Radiology,The Second Affiliated Hospital of Shandong First Medical University,Taian, Shandong,271000,China

^bDepartment of nuclear medicine, Qilu Hospital, Shandong University, Jinan, 250000,China

Eiman, Soliman. WLabib,G Eitantawi, etal(2012). Role of matrix metalloproteinase-3 (MMP-3) and magnetic resonance imaging of sacroiliitis in assessing disease activity in ankylosing spondylitis[J].Rheumatol Int,32:1711–1720.
Chen C, Lin K, Yu D, et al. Serum matrix metalloproteinases and tissue inhibitors of metalloproteinases in ankylosing spondylitis:MMP-3 is a reproducibly sensitive and specific biomarker of disease activity[J]. Rheumatology. 2006;45:414–20.
Eder L, Chandran V, Shen H, Cook RJ, Uladman DD et al. (2010).Is ASDAS better than BASDAI as a measure of disease activity in Rheum Dis,axial psoriatic arthritis?[J].Ann69:M,2160–2164.
Aydin SZ, Atagunduz P, Direskeneli H. Active disease requiring TNF-alpha- antagonist therapy can be well discriminated with different ASDAS sets:aprospective,follow-up of disease activity assessment in ankylosing spondylitis[J]. Clin Exp Rheumatol. 2010;28:752–5.
Khan MA, et al. Ankylosing spondylitis: introductory comments on its diagnosis and treatment[J]. Ann Rheum Dis. 2002;61(12):3–7.
Song IH, Sieper J, Rudwaldit M. Diagnosing early ankylosing spondylitis[J]. Curr Rheumatol Rep. 2007;9(5):367–74.
Unlu E, Pamuk ON, Cakir N. Color and duplex Doppler sonography to detect sacroiliitis and spinal inflammation in ankylosing spondylitis. Can this method reveal response to anti-tumor necrosis factor therapy[J]. Rheumatol. 2007;34(1):110–6.
Gaspersic N, Sersa I, Jevtic V, et al. Monitoring ankylosing spondylitis Therapyby dynamic contrast enhanced and diffusion weighted magnetic resonance imaging[J]. Skeletal Radiol. 2008;37:123–31.
Kumar V, Gu Y, Basu S, et al. Radiomics: the process and the challenges. Magn Reson Imaging. 2012;30(9):1234–48.
Mannil M, Eberhard M, Becke R, AS., (2017).Normative values for CT based texture analysis of vertebral bodies in dual X-ray absorptiometry-confirmed, normally mineralized subjects[J].Skeletal radiology,46(11):1541–1551.
Rudwaleit M. The development of Assessment of Spondyloarthritis international society classification criteria for axial Spondyloarthritis(partІ): classification of paper patients by expert opinion including uncertainty appraisal[J]. Ann Rheum Dis. 2009;68(6):770–6.
Rudwaleit M, et al. The development of Assessment of Spondyloarthritis international Society classification criteria for axial Spondyloarthritis(partП): validation and finalselection[J]. Ann Rheum Dis. 2009;68(6):777–83.
Zhao YH, Li SL, Liu ZY et al. (2015).Detection of active saeroiliitis with ankylosing spondylitis through intravoxel incoherent motion diffusion weighted MR imaging.Eur Radiol, 25 (9):2754–2763.
Machado PM, Landewé R, Heijde DV. Ankylosing spondylitis disease activity Score(ASDAS): 2018 update of the nomenclature for disease activity states. Ann Rheum Dis;. 2018;77:1539–40.
Maksymowych WP, Inman RD, Salonen D. Spondyloarthritis Research Consortium of Canada magnetic resonance imaging index for assessment of sacroiliac joint inflammation in ankylosing spondylitis. Arthritis Theum. 2005;53(5):703–9.
BentonN StewartN. CrabbeJ, et a1.(2004).MRI of the wrist in early Rheumatoid arthritis can beused to predict functional outcome at 6 years[J].AnnRheuxnDis,63(5):555–561.
Puhakka KB, Jurik AG, Schiottz-Christensen B et al. (2004).Magnetic Histopathologic evidence imaging of sacroiliitis in early seronegative spondylarthropathy.Abnormality correlated to clinical and laboratory findings [J].Rheumatology,43( 2):234–237.
Maksymowicz H. Kowalewski K,Lubkowska K,etal.(2010).Diagnostic value of gadolinium-enhanced MR imaging of active sacroiliitis in seronegative spondylo -arthropathy [J].Pol J Radiol,75:58–65.
Landewé RB, Hermann KG, van der Heijde DM, et al. Scoring sacroiliac joints by magnetic resonance imaging. multiple-reader Reliab experiment[J] Rheumatol. 2005;32:2050–5.
Kwee TC, Takahara T, Ochiai R, et al. Whole-body diffusion- weighted magnetic resonance imaging[J]. Eur J Radiol. 2009;70(3):409–17.
Ai F, et al. Value of diffusion-weighted magnetic resonance imaging in early diagnosis of ankylosing spondylitis[ J]. Rheumatol Int. 2012;32(12):4005–13.
Reischauer C, Patzwahl R, Koh DM, et al. Texture analysis of apparent diffusion coefficient maps for treatment response assessment in prostate cancer bone metastases: a pilot study[J]. Eur J Radiol. 2018;101:184–90.
Nardone V, Tini P, Croci S, et al. ).3D bone texture analysis as a potential predictor of radiation induced insufficiency fractures[J]. Quant Imaging Med Surg. 2018;8(1):14–24.
Valentinitsch A, Trebeschi S, Kaesmacher J, et al. Opportunistic osteoporosis screening in multidetector CT images via local classification oftextures[J]. Osteoporos Int. 2019;30(6):1275–85.
Fengxian Wang C, Chu, Li, Zhu et al. (2018).Whole-Lesion ADC Histogram Analysis and the Spondyloarthritis Research Consortium of Canada (SPARCC) MRI Index in Evaluating the Disease Activity of Ankylosing Spondylitis[J].Original Reserch,DOI: 10.1002/jmri.26568.
Scalco E. Rancatia T,Pirovano L,et a1.(2018).Texture analysis of T₁W and T₂W MR images allows evaluation of radiation-induced changes of internal obturator muscles radiotherapy for prostate cancer[J].Med Phys.45(4):1518–1528.

No competing interests reported.

Prediction of the Activity of Early Ankylosing Spondylitis using Radiomics Texture Analysis on STIR

Status:

Version 1

Abstract

Purpose

Methods

Results

Conclusion

Figures

Introduction

Materials And Methods

Patients

Specific Criteria For Clinical Evaluation Of Sacroiliac Arthritis

Mri Evaluation Of Sacroiliac Arthritis

Roi Segmentation

Radiomic Features Extraction And Establishment Of Omics Model

Statistical analysis

Results

Discussion

Abbreviations

Declarations

References

Additional Declarations

Status:

Version 1