This study was approved by the ethics committee of our university (approval no. ERB-110089 and ERB-110327).

Figure 1 shows a photograph and the design of the novel jawbone SPECT phantom, which was designed based on 3-dimension (3D) skull data (https://free3d.com/ja/3d-model/skull-v3--785914.html). This phantom was created by outsourcing (NMP Business Support Co., Ltd., Hyogo, Japan). The phantom consisted of a jawbone section and a cylindrical container. The jawbone section was made of epoxy resin, forming a horseshoe-shaped structure 12 mm thick, measuring 120 mm front to back and 100 mm left to right. Eight lesion areas 1 mL in size were set in the normal bone area of 39.63 mL. Technical evaluation of jawbone SPECT could be performed by varying the radioactivity concentration, location, and size of the lesion. The cylindrical outer had a diameter of 200 mm and a height of 185 mm. It was designed to match the size of the Brain Tumor phantom, a 11C-methionine brain tumor PET phantom, considering the combined dimensions of the neck and maxillofacial. The outer container of the novel jawbone SPECT phantom was filled with 3.0 kBq/mL of 99mTc solution, and the normal bone area was filled with 99mTc in bone-equivalent solution (99mTc: 39.0 kBq/mL and K2HPO4 solution: 0.50 g/cm3). K2HPO4 is a bone-equivalent solution with a composition comparable to that of cranium [23]. Figure 2 shows a CT image of the phantom. Simulated lesions of 1 mL were placed on both sides. The radioactivity ratios of the normal bone area to the lesions (lesion/normal ratio; LNR) were 1:2, 1:4, and 1:8 (lesion radioactivity: 78, 156, and 312 kBq/mL, respectively).

Photograph and design of the novel jawbone SPECT phantom. The phantom consists of a jawbone section and a cylindrical container. The jawbone section is a horseshoe-shaped structure, and eight lesion areas are set in the normal bone area

CT image of the novel phantom. The jawbone area was filled with a bone-equivalent solution (K2HPO4) with a concentration of 0.50 g/cm3, which results in a CT value comparable to that of the normal mandible

The concentrations of 99mTc and K2HPO4 solutions were determined by comparing clinical and phantom evaluations in a preliminary study. The radioactivity concentration of the 99mTc solution in the outer container was determined by placing a sufficiently large region of interests (ROIs) in the paracervical soft tissue and adopting the measured concentration. The concentration of the normal jawbone was determined by matching the measured radioactivity concentration of the normal side to that measured in the phantom. The measured radioactivity concentration ratio of the lesion to the normal area was 2.7 for the minimum, 4.2 for the median, and 7.9 for the maximum, so the LNR was set accordingly. The average CT value of the normal mandible was 530 Hounsfield units (H.U.), and the concentration of K2HPO4 solution was adjusted to obtain a similar CT value.

All SPECT data were acquired using a dual-head gamma camera (Symbia T6; Siemens AG, Erlangen, Germany) equipped with a low-energy high-resolution collimator. The system had a spatial resolution of 7.4 mm with 99mTc placed 10 cm from the collimator. Image reconstruction was performed using the Syngo MI Apps version VA50C (Siemens Healthcare Co., Ltd., Munich, Germany).

The projection data were obtained using the continuous mode through 360° of rotation at 90 angular views. Data acquisition was performed with a 128 × 128 matrix, zoom of 1.23 × , pixel size of 3.9 mm, 360° acquisition (90 directions, 4° steps), and 25 s/step. A 99mTc photo-peak window was set as a 20% energy window centered at 140 keV. A sub-window for scatter correction was set as a 7% energy window on the lower side of the photo-peak window. SPECT acquisition was performed in a circular orbit with a radius of 15 cm using an attachment for the brain. A low-dose CT scan was performed at 130 kV, with the quality reference set at 120 mA, auto exposure control for the tube current, rotation time of 0.6 s, slice thickness of 2.5 mm, and pitch value of 1.6.

The SPECT images were reconstructed by using the ordered-subset expectation maximization (OSEM) algorithm with resolution recovery, scatter correction, and attenuation correction, which used low-dose CT. The number of subsets was fixed at 10 and the number of iterations was varied from 1 to 15. The SPECT images were post-smoothed using a 3D Gaussian filter (no filter, full width at half maximum (FWHM) of 3.90, 5.85, and 7.80 mm).

To investigate the basic properties of the reconstruction parameters, percent contrast (%contrast) and absolute recovery coefficients (ARCs) were calculated. %contrast and ARC were used as evaluation indices to assess optimal image reconstruction parameters in terms of image quality and quantitative accuracy. We placed ROIs with a 5 mm diameter on the lesion and normal bone area on the CT images and then copied the ROIs on the SPECT images. The %contrast and ARC were calculated as follows:

$$= \left(\frac-1\right)/\left(\frac-1\right)\times 100\left(\%\right)$$

(1)

$$\text=\text\times (Cl\textCn)/(Al\textAn)\times 100\left(\right)$$

(2)

Here, Cl and Cn are the average counts in the ROI in the lesion and normal bone, respectively. Al and An are the radioactivity in the lesion and normal bone area, respectively. The cross-calibration factor (CCF) converts count values [counts/pixel] to radioactivity concentration [Bq/mL]. The CCF was obtained from the correlation between radioactivity and counts per second of a cylinder phantom. The cylindrical phantom had a diameter of 160 mm and a height of 150 mm. The CCFs were calculated and applied for each reconstruction parameter.

We retrospectively enrolled 19 MRONJ patients who underwent bone SPECT examination at our hospital from April 2022 to February 2023. MRONJ diagnosis and clinical stage were confirmed according to the 2022 the American Association of Oral and Maxillofacial Surgeons definition [24]. The clinical stages are defined based on the presence of infection as determined by clinical findings. The relationship between stage and clinical symptoms is shown in Table 1. The patient characteristics are shown in Table 2. The patients included 7 men and 12 women with a mean age ± standard deviation of 72.0 ± 10.8 years (range: 41–86 years). The MRONJ stages of the patients were as follows: stage 1 (n = 6), stage 2 (n = 10), and stage 3 (n = 3). The patients’ target illnesses were as follows: cancer (n = 10), osteoporosis (n = 8), and rheumatoid arthritis (n = 1).

In all patients, 99mTc-hydroxymethylene diphosphonate was intravenously injected and data acquisition was initiated approximately 4 h later. SPECT acquisition was performed after planar acquisition. The injected dose, calculated by subtracting the post-injection remaining radioactivity in the tube and syringe from the pre-injection radioactivity, was 952 ± 137 MBq. The uptake time for SPECT acquisition was 250.3 ± 7.9 min.

The SPECT/CT scanner, SPECT acquisition conditions, and CT imaging conditions were the same as for the phantom study. Immediately after data acquisition, a low-dose CT scan was performed. SPECT images were reconstructed using two sets of parameters: OSEMjaw, with reconstruction parameters determined by the results of the phantom study; and OSEMcurrent, with a subset of 10, number of iterations of 6, and FWHM of 7.80 mm for the Gaussian filter, which were the image reconstruction conditions determined previously using a spherical phantom [25].

SUVmean values of the clinical images were calculated by using GI-BONE software (AZE Co., Ltd., Tokyo, Japan). Volumes of interest were placed in the lesion areas. The average SUV that was 40% or more of SUVmax in the volume of interest was defined as SUVmean. SUV was calculated by

$$\text = \frac/\text}/\text}$$

(3)

The quality of the SPECT images was evaluated visually by a board-certified nuclear medicine physician, a board-certified radiologist, an oral and maxillofacial surgeon, and a board-certified nuclear medicine radiological technologist. The score was graded from 1 to 5 (1, very poor; 2, poor; 3, moderate; 4, good; and 5, excellent) in terms of detectability of lesions, including the contrast and the sharpness of the lesion area.

SUVs were compared between clinical stages using the Kruskal–Wallis test, and subsequent post hoc analysis using the Steel–Dwass test. The visual score was acquired by averaging the results of the observers’ scores. The visual scores for two parameters were compared by using Wilcoxon signed rank tests. Differences were considered statistically significant when p < 0.05.