DXA Regions of Interest (ROI)

Overview 

Dual-energy X-ray absorptiometry (DXA) provides quantitative measures of body composition and bone mineral density (BMD) and is part of the imaging project within UK Biobank. ROI boundaries are automatically set by GE Lunar’s enCORE software at the time of the scan. If needed, these positions can be adjusted by the user (radiographer) during scan analysis within the enCORE software. This article explains how DXA imaging is performed and details the ROIs defined on each scan.

This data is available in several categories, primarily under:

Equipment and acquisition

UK Biobank DXA images are acquired using GE Lunar iDXA scanners. Each scanner is operated using enCORE software (GE Healthcare) which uses well-defined ROIs generated automatically by GE software.

DXA modalities

UK Biobank includes DXA scans on 5 areas:

1. Whole-body
2. Knees
3. Hips
4. Lumbar spine (L1-L4)
5. Lateral vertebral assessment (LVA)

DXA scans of the whole body, lumbar spine (AP) and hips are analysed by radiographers at, or soon after, acquisition using GE Lunar enCORE software. This analysis generated the numerical measures of body composition and bone mass in Categories 124 and 125.

Most variables in Categories 124 and 125 are outputs generated directly by the GE Lunar enCORE software during scan analysis. However, Category 125 also includes some fields that were derived and returned by researchers, such as Field 20310 and Field 20318. These were generated outside the enCORE software and integrated into this category.

DXA images are also acquired for the knee and lateral vertebral assessment (LVA). These images were not analysed and no derived measures are provided for these scans.

Raw images from all DXA scan types are provided without further analysis for potential future use by researchers under Field 20158. As ROIs are defined in the proprietary software, manual image re-analysis of these images is limited without access to GE Lunar enCORE software.

For more information on the DXA procedure, please see Resource 502.

1. Whole-body composition scan

Figure 1. Whole-body DXA image showing ROIs segmentation of head, arms, torso, spine, pelvis, and legs. Blue lines indicate anatomical segmentation used to define core ROIs. The pink lines indicate special ROIs, such as the android and gynoid.

2. Hip scan (left and right hip)

Figure 2. DXA hip ROIs scan. Blue lines indicate anatomical segmentation used to define hip ROIs. The ROIs correspond to femoral regions, such as the lower and upper femoral neck, trochanter, Ward’s, shaft, and total hip (combined region of the femoral neck, trochanter, and shaft regions).

3. Lumbar spine (L1-L4)

Figure 3. Lumbar-spine DXA scan (anteroposterior) showing ROIs over vertebrae L1-L4. Where (1) shows the intervertebral (IV) markers are placed between the vertebral bodies. (2) is used to identify the lowest point of bone density. Blue lines indicate core anatomical segmentation used to define vertebrae.

4. Lateral vertebral assessment (LVA)

Spanning T4-L4 in a lateral projection.

5. Knee (left and right)

Knee images are for assessment of the tibiofemoral joint space and no derived measures were generated at the time of scan.

 

Pixel Size in DXA DICOM files

Unlike DXA, imaging modalities such as MRI typically define voxel dimensions explicitly at acquisition and store this information in the DICOM metadata. DXA images, by contrast, are reconstructed onto a 2D grid, meaning the pixel spacing reflects a derived image representation rather than a direct sampling resolution.

In the UK Biobank DXA dataset, the DICOM attribute Pixel Spacing (0028,0030) is not populated in the image headers, and pixel spacing must therefore be inferred indirectly. Several studies have reported effective pixel spacing values of approximately 0.25–0.30 mm (e.g., Hind et al., 2015; Farzi et al., 2020), although these publications do not explicitly describe the derivation method.

If a pixel size estimate is required for analysis, one way you can derive pixel spacing is from the relationship between the image matrix dimensions and the corresponding physically scanned area.

Exposed Area

The exposed area is the physical field of view that the DXA scanner irradiates during acquisition. It will vary in size depending on factors like the size of the participant or the body part being scanned.

Example: A hip scan might have an exposed area of 170 mm width and 160 mm height. These values come directly from the DICOM header in Exposed Area parameter.

Pixel Matrix or Number of Pixels

A DICOM image consists of an image matrix. This is a 2D grid defined by Rows (height) and Columns (width)

Example: The DXA image for the hip scan may have Rows = 640 and Columns = 680 . This means the pixel matrix is 680 × 640.

• These values are present in the DICOM header.

Computing Pixel Size

pixel_size_width_mm  = Exposed width (mm)  / Number of columns
pixel_size_height_mm = Exposed height (mm) / Number of rows

References:

Farzi, M., Pozo, J. M., McCloskey, E., Eastell, R., Harvey, N., Wilkinson, J. M., & Frangi, A. F. (2020). A Spatio-Temporal Ageing Atlas of the Proximal Femur. IEEE Transactions on Medical Imaging, 39(5), 1359–1368. https://doi.org/10.1109/TMI.2019.2945219

Hind, K., Cooper, W., Oldroyd, B., Davies, A., & Rhodes, L. (2015). A Cross-Calibration Study of the GE-Lunar iDXA and Prodigy for the Assessment of Lumbar Spine and Total Hip Bone Parameters via Three Statistical Methods. Journal of Clinical Densitometry, 18(1), 86–92. https://doi.org/10.1016/j.jocd.2013.09.011