Validity of a computer-assisted manual segmentation software to quantify wrist erosion volume using computed tomography scans in rheumatoid arthritis

The optimization of imaging measures is an important strategy for evaluating and monitoring bone damage in rheumatoid arthritis (RA)[1]. Although conventional radiography (CR) remains the cornerstone of imaging modalities in RA, computed tomography (CT) appear more sensitive for the detection of bone erosions in comparison to CR, magnetic resonance imaging (MRI) and ultrasonography (US). For that reason, CT can be considered the standard reference method for the detection of erosive bone destructions in early stage of the disease[2‐6]. Current generation of ultrafast CTs allows acquiring high-resolution volumetric data in few seconds and providing detailed anatomical informations. In particular, 3-D volume rendering techniques makes feasible to generate high-quality images, offering a realistic anatomical view of the body and organs from tomographic data.

Radiographic scores, such as the Larsen[7], Sharp scores and its modified versions[8, 9], are standard methods to evaluate and monitor the joint damage in RA[10, 11]. Despite considerable effort to either reduce or at least define the intrinsic limitations of radiographic scores, some problems remains, especially for the wrist’s evaluation, such as reader variability, floor and ceiling effects[12‐14] and inability to accurately quantify damage and its progression. To circumvent these problems, the feasibility of a computerized image analysis of erosion volumes of the hands has been evaluated. The computer system has been shown to provide reproducible data, but showed poor correlation with common standard scores used to define bone erosions[15‐18]. Unlike the CR, no method of qualitative score was developed specifically for CT images. Therefore, monitoring changes of the erosions volume is an important task for a correct evaluation of disease progression and response to treatment. Measurements of erosions volume by CT images has been described by Duryea et al.[19] and Døhn et al.[20]. In particular, Duryea et al.[19] have developed a semi-automated method for the evaluation of the profile of carpal bones and subsequent determination of the volume erosion by comparing the images obtained at baseline and subsequent checks.

The objectives of our study were to investigate the performance of the CR for the detection of bone erosions in the wrist of patients with RA, using multidetector CT as the reference method and to assess the validity of a computer-assisted manual segmentation software to quantify the erosion volume of wrist detected by CT.

A total of 375 wrist bones were assessed for erosions in RA patients and 90 wrist bones in healthy controls. All patients had at least one erosion on CT, while none was seen in healthy controls. CT and CR detected 145 and 37 erosions, respectively. All erosions detected by CR were confirmed by CT. The lunate, trapezium, capitate and the first metacarpal head bones were common sites of erosions (Figure 4). Considering CT as the reference method, the sensitivity, specificity and overall accuracy of CR for detecting erosions were 25.5%, 98.3% and 70.1%, respectively. Pattern differences between early and established disease were not detected.

The total erosion volume in each subject ranged from 0.001 cm³ to 2.01 cm³. The highest erosion volumes were noted in the capitate (highest value 0.42 cm³) and lunate (highest value 0.26 cm³) bones. Spearman’s RAMRIS CT score of individual wrist bones in all patients, correlated with the erosion volume on CT (rho 0.823; p < 0.0001) and with the ratio between erosion volume and the corresponding bone volume on a percentage basis (rho = 0.914; p < 0.0001). The total Sharp/van der Heijde erosion score of the all wrist bones was slight correlated with the RAMRIS CT score (rho 0.143; p = 0.008), but not with total volume erosion measurement on CT (rho 0.098; P = 0.069). The ICC for manual segmentation showed very high precision (ICC = 0.901; 95% CI 0.882 to 0.931), with no significant difference in the inter-rater reproducibility. The mean and 95% CI of the absolute difference between the volume measurements were 0.158 (95% CI for the mean 0.105 to 0.211), and 0.165 (95% CI for the mean 0.1081 to 0.222) cm³, respectively (difference 0.00924; 95% of difference −0.069 to 0.088), which are depicted graphically in the Bland–Altman difference plot of Figure 5.

Considering the wrist frequently involved in RA[28, 29] and that anatomical damage at this level has a predictive significance of evolution of the disease, its evaluation assumes an important clinical significance[28‐32]. The lunate, trapezium, capitate and the first metacarpal head bones are common sites of erosions. Østergaard et al[33] found that capitate, ulna, lunate, triquetrum, and scaphoid bones were the most frequently bone involved in a MRI datasets from 258 RA patients (126 early RA patients).

Experiences from comparative studies of CR, US, MRI and CT have shown that CT is the most sensitive imagine technique for the detection of bone erosions[2, 3, 34‐38] and, therefore, can be considered the gold standard for the evaluation of wrist anatomical damage in RA. Difficulties of CR for a detail evaluation of wrist joints are mainly due to a projectional superimposition of bones in this complicated anatomical area, high irregularities of the bone margins (e.g. at level of ligaments attachment) and the presence of nutritive foramina that can appear like erosions. These aspects make difficult the discrimination between normal anatomy and presence of erosions.

Measurements of erosion progression in patients with RA have generally relied upon semiquantitative scoring systems[7, 11, 14]. Despite considerable effort to either reduce or at least define the intrinsic limitations of radiographic scores, problems remain with reader variability and floor and ceiling effects[12‐14]. Principles of OMERACT RA MRI system[23, 24] were applied to score bone erosions by CT. Despite a scoring system for CT images is not yet available, we found OMERACT RA MRI system reasonable to use due to the similar tomographic visualization of joints on CT and MRI, the good inter-intraobserver reliability and a high level of sensitivity and specificity demonstrated by RAMRIS system in monitoring joint destruction in RA[39‐41].

Recently, technological advances in CT imaging modalities and the introduction of 3D image reconstruction through computer systems, allowed the visualization of tiny defects of the wrist a feasible task[42]. High resolution CT systems are fast and produce high spatial resolution images with near isotropic voxels. Therefore, 3D image reconstruction of segmented carpals is significantly more accurate. The visualization and processing of 3D data require special navigation tools and multidimensional rendering software that are available on high end 3D rendering workstations that are most accessible in academic and specialized imaging centers.

The successful diffusion of OsiriX in the medical community[43] is mainly due to the fact that it is a cost effective alternative to not-free softwares and can be customized to match the needs and specific usage in clinical setups[26]. Previously published studies have investigated segmentation of the carpal bones of wrist by CT scans with a variety of goals. Snel et al.[44, 45] and Sebastian et al.[46] performed wrist segmentation to examine the kinematics of the carpal bones. Other studies have examined wrist scans as a validation of the general 3D segmentation methods. Van Cleynenbreugel et al.[47] and Yao et al.[48] described a semiautomatic method to segment bones on spiral CT scans. These techniques were developed to better understand carpal kinematics or as part of a more general 3D segmentation package. As the same Døhn et al.[49] and Bird et al.[50], we found that the volume measurements of erosions by CT was highly reproducible and closely correlated with the semiquantitative scores of bone erosions according to OMERACT RAMRIS system score. This aspect support the evidence for the semiquantitative MRI measures of erosion as a valid measure of bone destruction in RA. Although the volume measurement of erosion by CT has proved to yield reproducible findings, it showed poor correlation with joint space narrowing score[15] and with standard erosion score[18]. Since CT scanning appears more accurate than the observation of 2D images, this result can compromise the validity of the total Sharp/van der Heijde erosion score. However, the disadvantages of CT scanning are due to its relatively higher cost and risk of hazard from radiation exposure[51, 52]. Further, our method evaluates 15 carpal sites in one shot, each of them potentially containing multiple erosions. A limited number of joints can be examined by using CT due to the exposure to ionizing radiation. Consequently, a smaller number of sites can be assessed by CT which potentially lead to a risk of missing signs of erosive progression of the disease. Additionally, the method could also be developed to measure erosion volumes in the proximal interphalangeal and metacarpophalangeal joints that are common sites of erosions as well.

The number of erosions detected on CT indicates that this imaging modality is high sensitive for detecting bone erosions in wrist of RA patients. The close correlation with erosion volumes determined by CT provides further evidence for the semiquantitative OMERACT RAMRIS erosion score as a valid tool of anatomical bone damage evaluation. However, the sensitivity to change is not yet established and the disadvantages of CT include the necessity for ionising radiation and inability to visualize soft tissue. Moreover, the computer-assisted manual segmentation technique is complicated and time-consuming. The longer time for the volume calculations was largely due to the manual nature of the segmentation process as well as to the large number of bones assessed at level of wrist. Whether computer-assisted manual segmentation technique is sufficiently precise the evaluation of responsiveness has not yet been tested in our study and, therefore, it will be necessary to evaluate this aspect in further investigations on larger prospective studies.