Quantitative Analysis of Multimodal Skeletal SPECT/CT Reconstructions in Diagnosing Medication-related Osteonecrosis of the Jaw

Abstract

Abstract Aim Our goal was to assess visual and quantitative aspects of multimodal skeletal SPECT/CT reconstructions (recon) in differentiating necrotic and healthy bone of patients with suspected MRONJ. Methods Prior to surgery, 20 patients with suspected MRONJ underwent SPECT/CT of the jaw 3–4 hours after injection of Tc-99m-DPD (622±112.4 MBq). SPECT/CT data were reconstructed using the multimodal xSPECT Bone and xSPECT Quant algorithms as well as the OSEM-algorithm FLASH 3D. For analysis, we divided the jaw into 12 separate regions. Both xSPECT Bone and FLASH 3D datasets were scored on a four-point scale (VIS xSPECT; VIS F3D), based on the intensity of localized tracer uptake. In F3D and xSPECT Quant datasets, local tracer uptake of each region was recorded as semi-quantitative uptake ratio (SQR F3D) or SUVs, respectively. ROC analysis was performed. Postoperative histologic results served as gold standard. Results VIS F3D, VIS xSPECT and SQR F3D did not differ significantly in diagnostic accuracy (VIS xSPECT sensitivity=0.64; specificity=0.89). Of the quantitative parameters, SUVpeak yielded the best interobserver reproducibility. SUVpeak was 9.9±7.1 (95%CI: 7.84–11.95) in MRONJ regions, as opposed 3.6±1.8 (95% CI:3.36–3.88) elsewhere, with a cutpoint of 4.5 (sensitivity=0.83; specificity=0.80). Absolute quantitation significantly surpassed VIS and SQR (p<0.05) in accuracy and interobserver agreement (SUVpeak: κ=0.92; VIS xSPECT: κ=0.61; SQR F3D κ=0.66). Conclusion Absolute quantitation proved significantly more accurate than visual and semi-quantitative assessment in diagnosing MRONJ, with higher interobserver agreement.