Validation of resolution recovery CT-attenuation corrected SPECT/CT MIBI parathyroid imaging

Objectives This investigation evaluated the degree to which ^99mTc-MIBI SPECT/CT aids in detecting & localizing parathyroid lesions in pts. with primary hyperparathyroidism (PHP) when SPECT data are reconstructed using commercially available OSEM resolution recovery & CT-attenuation corrected reconstruction algorithms designed for bone imaging, & not validated for parathyroid imaging.

Methods Data for 202 pts. with PHP were analyzed. Analyses were limited to pts. with surgically confirmed single gland disease to minimize variables. All pts had surgically confirmed lesion locations & wts. 2 experts reviewed SPECT/CT images independently in 2 separate reading sessions. Readers graded lesion detection confidence from 0=definitely normal to 4=definitely abnormal. ROC curves were generated with surgical findings as reference standard. Chi-squared analysis compared test sensitivity, specificity & accuracy against previously reported results (Radiology 2008;248:221-32) for 409 pts of a control group, for whom SPECT data were reconstructed by conventional filtered backprojection (FB).

Results Mean wt. for the 202 lesions was 883±948 mg, similar to the control group (p=0.33). Male:female ratios & pt. ages were similar in test & control groups (p=0.13, p=0.40, respectively). Optimal accuracy (ROC area) for the 4 readings by the 2 readers of 95±2% was obtained using a discrimination threshold>0. Comparing dichotomized OSEM readings in the test group to FB readings in the control group, sensitivities were similar (92% vs. 90%, p=0.62). Specificity (93% vs. 84%, p<0.0001) & accuracy (92% vs. 87%, p=0.004) were significantly higher in the test group. There was “good agreement” for intra-rater reproducibility (Κ= 0.79) & inter-rater agreement (Κ=0.73).

Conclusions OSEM with resolution recovery & CT-attenuation correction is highly accurate for preoperative parathyroid lesion detection in pts. with PHP with sensitivity similar, & specificity superior to conventional filtered backprojection processing