Assessment of Bone Fracture Healing Using Micro-Computed Tomography

Micro-CT is a very common approach for quantifying 3D morphology and quality of bone. Although analysis protocols are fairly established for characterizing intact cortical and trabecular bone, there is a less consensus on the protocol for analyzing fracture healing. This technique involves noninvasive imaging and accurate, calibrated 3D analysis using a good balance of manual and automated techniques.

It also uses a sophisticated and flexible software environment. Demonstrating the procedure will be Hwabok Wee, a research associate from my laboratory. To begin, take a custom-developed 3D printed scanning fixture or similar that includes a miniature hydroxyapatite phantom for bone mineral density calibration.

Place up to six long bone samples into the fixture for the simultaneous scanning of multiple samples. Place the prepared fixture in a syringe or a conical tube similar to the scanning field of view's diameter. Fill the syringe with a preservative, like saline, to prevent samples from drying out during the scanning process.

After confirming the calibration of the micro-CT machine, align the sample fixture's center line with the approximate center line of the micro-CT to ensure that the samples are within the field of view and their long axes have orientation approximately coinciding with the axial direction of the resulting images. Next, set the scanning parameters of the micro-CT system by setting the energy or intensity to 55 kilovoltage peak, current to 145 microampere, isotropic voxel size to 10.5 micrometers, and integration time to 300 milliseconds. Then visually inspect the scout images in different views to cover the whole volume of all callus samples.

Start the scan acquisition, and, once complete, convert the images to a DICOM stack to import them into the analysis software. To start with image cropping, pick one sample at a time, and crop each image stack, ensuring the whole sample is included in the cropped volume. Save the cropped image by clicking the File tab on the top left side of the screen, selecting Save Project As, and then selecting Minimize project size.

To denoise the image, click the File tab and choose the image to be processed using Open Data, which opens the image in the Project View window in the top left corner of the screen. Right click to select Image Processing, followed by Filter Sandbox, and then click Create. In the Properties window on the left bottom corner of the screen, choose Data as the preview type.

Select the filter type from the dropdown menu next to Filter and choose 3D for interpretation. Keeping Separable as the kernel type in the dropdown menu, fill in the values of standard deviation and kernel size factor in the available empty box. Then select Same as input from the dropdown menu next to the output.

And finally, click Apply. For misaligned samples, the user may perform image realignment by creating a 3D-rendered image of the sample by selecting the filtered, cropped image from the Project View window. Right click to select Display and then Volume Rendering from the dropdown menu.

Then click Create to visually check the 3D-rendered image in the sagittal and frontal planes. Then manually rotate the rendered volume to obtain a good alignment in the longitudinal axis. To apply the transformation in the rotated images, in the Properties window, click the Transform Editor.

Then go to Transform Editor Manipulator and select Transformer from the dropdown menu. If needed, rotate, realign, and then click the Transform Editor again to lock the image. Next, to create new transverse plane image slices, resample the filtered image by selecting the image from the Project View window.

Right click to select Geometry Transform, followed by Resample Transformed Image from the dropdown menu, and then click Create. In the Properties window, go to Data, and from the dropdown menu, select Standard for interpolation, choose Extended for mode and Voxel Size for preserve. Enter zero in the available blank box for padding value, and finally, click Apply.

To define the volume of interest, go through the transverse image slices, identify the fracture callus's center plane, and define it based on the proximal and distal lens of the callus. If the callus ends are difficult to specify, define the volume based on a standardized distance away from the callus center plane. To segment the outer boundary of the callus following the reassembly of transformed images, click the Segmentation tab in the second tab row from the top of the screen.

In the segmentation editor window, select transformed image from the dropdown menu next to the image. In the MATERIALS window, double click Add. By doing so, two tabs named Material3 and Material4 will appear.

Right click to rename Material3 to callus and Material4 to cortical bone. Next, in the SELECTION window, click on the lasso icon. From the appearing options, select Freehand for the 2D mode, Inside for the 3D mode, and both Auto trace and Trace edges for options.

Then use the lasso to mark the outer boundary of callus. Repeat the contouring steps with slices sampled across the volume of interest, and the slices can be spaced apart, for example, by 20 slices. To create a complete callus label, in the MATERIALS window, choose the callus file, click the Selection tab on the top of the screen, and select Interpolate from the dropdown menu.

Then, in the Selection window, click on the plus sign. Next, segment the cortical bone, including the medullary cavity. Then interpolate the contoured periosteal cortex surface to create a cortical bone label as performed previously for the callus.

To calculate the callus's contoured volume and mean gray value, click the Segmentation tab on the top row of the screen and select Material Statistics from the dropdown menu to generate a table of calculated values. Note that the values of the cortical bone and the callus are provided separately after subtracting the cortical bone. Export the generated table and save the data by clicking Export into Workspace.

To convert the grayscale units to bone mineral density, crop the 3D image of the 4.5-millimeter HA phantom from the whole image and click Segmentation. To draw circles at the first and the last slices, in the MATERIALS window, click Add four times. Then right click to rename Material3, 4, 5, and 6 to Phantom1, 2, 3, and 4 respectively.

Next, select Phantom1. Click the brush icon in the Selection window and use the slider to adjust the brush size such that the size of the circle is smaller than that of the phantom. To create a volume for each HA cylinder, apply interpolation by selecting Phantom1 in the MATERIALS window, clicking the Selection tab on the top row of the screen, and selecting Interpolate from the dropdown menu.

Then, in the SELECTION window, click the plus sign. Repeat this process with the remaining HA cylinders. Use the generated 3D labels to calculate the mean gray values of the four analyzed HA cylinders.

Plot the mean gray values and the corresponding bone mineral density, or BMD, values provided by the phantom manufacturer. Generate a correlation equation between BMD and the gray values using linear regression. To segment the mineralized callus, click Add in the MATERIALS window.

Then right click to rename the new material to callus mineralized. Next, select Callus in the MATERIALS window, followed by clicking Select. Then click Threshold in the SELECTION window.

Select the lower value and apply the calculated threshold from the HA phantom in threshold masking. Then click Select only current material in the SELECTION window, followed by clicking Select Masked Voxels. Then select callus mineralized in the MATERIALS window and click the plus sign in the SELECTION window.

Finally, click 3D of callus mineralized in the MATERIALS window. The micro-CT images analyzed at three time points showed the substantial formation of mineralized callus on day 14. Incremental increases in bone fraction volume and bone mineral density were seen as healing proceeded from day 14 to days 21 and 28, consistent with bony bridging of the fracture gap.

As expected, the callus underwent resorption or remodeling between days 21 and 28, as evidenced by a decline in total callus volume. Cortical bridging of the callus was more evident at day 28 than at any preceding time point. For complex fractures, we recommend carefully reviewing the resulting semi-automated segmentation by scrolling through all image slices, sometimes in different planes, and adjusting contours if needed.

Once the fracture callus is accurately segmented, outcome parameters, in addition to the ones reported in this protocol, can be computed with additional scripts. These include, for example, moments of inertia and connectivity density.