Improving the accuracy of bone mineral density using a multisource CBCT

Abstract

Multisource cone beam computed tomography CBCT (ms-CBCT) has been shown to overcome some of the inherent limitations of a conventional CBCT. The purpose of this study was to evaluate the accuracy of ms-CBCT for measuring the bone mineral density (BMD) of mandible and maxilla compared to the conventional CBCT. The values measured from a multi-detector CT (MDCT) were used as substitutes for the ground truth. An anthropomorphic adult skull and tissue equivalent head phantom and a homemade calibration phantom containing inserts with varying densities of calcium hydroxyapatite were imaged using the ms-CBCT, the ms-CBCT operating in the conventional single source CBCT mode, and two clinical CBCT scanners at similar imaging doses; and a clinical MDCT. The images of the anthropomorphic head phantom were reconstructed and registered, and the cortical and cancellous bones of the mandible and the maxilla were segmented. The measured CT Hounsfield Unit (HU) and Greyscale Value (GV) at multiple region-of-interests were converted to the BMD using scanner-specific calibration functions. The results from the various CBCT scanners were compared to that from the MDCT. Statistical analysis showed a significant improvement in the agreement between the ms-CBCT and MDCT compared to that between the CBCT and MDCT.

Figure 1

The design of the ms-CBCT. (a) A schematic of the ms-CBCT scanner. (b) A photo of the CNT x-ray source array and the external multisource collimator.

Figure 2

(a) The anthropomorphic RANDO phantom consisting of multiple slabs with finite air gaps in between; (b) The home-made calibration phantom consisting of a 16 cm diameter SolidWater cylinder with 4 wells hosting removable inserts.

Figure 3

Axial images of the RANDO phantom from the MDCT (a) and CBCTs (b–d) at 110 kVp (window level: 600 HU, window width: 2800 HU). (b) ms-CBCT, (c) CBCT-1, (d) CBCT-N. The dark band is caused by the air gap between the multiple slabs of the RANDO phantom.

Figure 4

Sagittal images of the RANDO phantom from MDCT (a) and three CBCT scanners (b–d) at 110 kVp (window level: 600 HU, window width: 2800 HU). (b) ms-CBCT, (c) CBCT-1, (d) CBCT-N. The dark horizontal lines are from the air gap between the individual slabs of the RANDO phantom.

Figure 5

The RANDO phantom imaged using the ms-CBCT, after registration with the image from the MDCT and segmentation. (a,b) Segmentation of mandible , with yellow indicating the cortical bone and blue representing the cancellous bone. (c,d) Segmentation of maxilla.

Figure 6

The BMD values of the mandible and maxilla derived from the CBCT versus the values from the MDCT at 110 kV (a–c) and 90 kV (d–f). (a) ms-CBCT vs MDCT at 110 kVp; (b) CBCT-1 vs MDCT at 110 kVp; (c) CBCT-B vs MDCT at 110 kVp; (d) ms-CBCT vs MDCT at 90 kVp; (e) CBCT-1 vs MDCT at 90 kVp; (f) CBCT-M vs MDCT at 90 kVp. The red solid line represents the linear regression fit for the data, while the black dashed line denotes the line of identity (y = x), serving as a reference.

Figure 7

The Bland–Altman plots between the MDCT and ms-CBCT, CBCT-1, CBCT-N, and CBCT-M images. (a–c) CBCT images at the mandible and maxilla under the tube voltage of 110 kVp, (d–f) CBCT images at the mandible and maxilla under t the tube voltage of 90 kVp. The solid line represents the median difference (bias), and the dashed lines represent the upper (+ 95%LoA) and lower (− 95%LoA) 95% limits of agreement (LoA), which were estimated using the 2.5th and 97.5th percentiles due to non-normality of the data.