Background Bone erosions are a central feature of rheumatoid arthritis (RA). In particular small erosions not yet detectable on radiographs or even MRI, but already well depicted in high-resolution quantitative computed tomography (HR-pQCT) are a sign for early RA and may guide intervention before major bone destruction and functional impairment.
Objectives To compare standard segmentation tools implemented on the HR-pQCT scanner (XCT) with MIAF (Medical Image Analysis Framework) software in their ability to quantify erosion shape and size in metacarpophalangeal (MCP) joints. MIAF runs on external PCs and requires DICOM export of XtremeCT images.
Methods 2D threshold based contouring of bone is a standard feature of the XtremeCT software to segment the distal radius and tibia. These contouring algorithms along with the manual editing options were applied to segment erosions of MCP joints. In contrast MIAF uses more sophisticated 3D algorithms to separately segment periosteal bone surface and erosions. In particular the cortical break and (for larger erosions) the border between bone marrow and erosion can be better defined. HR-pQCT scans of the 2nd and 3rd MCP joints of ACPA positive RA patients, which represent a homogeneous subgroup with high level structural damage, were analyzed. Based on volume, erosions were categorized into 2 groups: <10mm3 (MIAF measurement) and >10mm3. Based on their shape, they also were visually classified into “regular” or “Irregular”. This classification was compared with sphericity (0–1), which can be obtained with MIAF but not with XCT.
Results 76 erosions of 65 (46F; 19M) patients were analyzed. Irregular erosions were larger (>10mm3) than regular erosions (χ(1)2=20.52, p<0.001). Irregular erosions had lower sphericity (0.53±0.08); than regular ones (0.63±0.09, t(74)= -4.645, p<0.001). Irregular erosions were considered smaller in MIAF than XCT (49.9±70.7 vs. 52.2±79.4; p=0.013), while regular erosions were larger than for XCT (7.0±9.4 vs. 4.0±4.5; p=0.002). There was a 47% absolute difference in size between MIAF and XCT. The differences could be largely explained over-segmentation of irregular erosions and under-segmentation of regular erosions in the XCT. The main reason was the under-segmentation of the cortical break.
Conclusions Standard tools implemented on XtremeCT scanners can be applied for measurement of erosion size but compared to MIAF the segmentation of the cortical break is less accurate resulting in differences in the determination of erosion volume. MIAF performs a separate bone and erosion segmentation, avoiding under/overestimating volume. The visual classification of erosions according to their shape could be useful to infer the amount of bone damage in RA patients but an operator independent quantitative parameter such as sphericity may be preferable as precision will likely increase.
Disclosure of Interest None declared