Assessing Spatial Concordance Between Theranostic Pairs Using Phantom and Patient-Specific Acceptance Criteria: Application to 99mTc-MAA SPECT/90Y-Microsphere PET

Abstract

Purpose: Predictive 3-dimensional dosimetry requires spatial concordance between diagnostic and therapeutic activity distributions. We assess similarity between theranostic pairs (^99mTc-macroaggregated albumin [MAA] single photon emission computed tomography [SPECT] and ⁹⁰Y microsphere positron emission tomography [PET]) in patients using criteria that account for spatial resolution differences and misregistration.

Methods and materials: Phantom-based acceptance criteria were determined using a liver phantom filled with ^99mTc and ⁹⁰YCl₃ and scanned with SPECT/computed tomography [CT] and PET/CT, respectively. Gaussian blurring was applied to PET to match ^99mTc phantom scan image quality. After rigid registration between SPECT/CT and PET/CT, perturbations up to ±3 voxels were applied to determine the similarity metric (SM) sensitivity. ^99mTc-MAA SPECT/CT and ⁹⁰Y microsphere PET/CT image pairs/patients (n = 23) were processed analogously. SMs calculated included the Pearson correlation coefficient (ρ_r), Lin's concordance correlation coefficient (ρ_c), Spearman's rank correlation coefficient (ρ_s), the mean squared difference, and the Dice similarity coefficient (DSC). Patient-specific acceptance criteria were determined by evaluating the SMs of the blurred PET compared with itself misregistered.

Results: After transforming PET to SPECT resolution, high similarity was found in phantom, with ρ_c, ρ_r, ρ_s > 0.98 ± 0.01, a mean squared difference of (4.1 ± 0.3) × 10^-4 and DSC > 0.85 ± 0.01 for investigated thresholds (5%, 30%, and 50%). SMs for patients varied from poor to good. A small percentage (13%-30%) of patient scans were acceptable using phantom-based acceptance criteria. The percentage increased slightly (17%-35%) using patient-specific acceptance criteria. DSC for most patients were substantially lower (average 0.95 vs 0.61 for 5% threshold) than phantom values.

Conclusions: At best, 35% of patients had an SM within the acceptance criteria established to account for imaging-related effects impacting spatial concordance between ^99mTc-MAA SPECT and ⁹⁰Y PET. Additional clinical factors should be evaluated in the future. The procedure of accounting for image-related effects when assessing spatial concordance can be applied to other theranostic pairs.

Fig. 1.

Diagram describing the processing of single photon emission computed tomography/positron emission tomography image pairs for phantom and patients. Green boxes represent operations, and blue boxes represent data.

Fig. 2.

Comparison of single photon emission computed tomography and PET phantom images with blurring and registration perturbations. (a) Images and scatter plot of voxel values without blurring the PET image. (b) Images and scatter plot of voxel values with a 10-mm blurring applied to PET image. (c) Scatter plot after perturbing the registration between the blurred PET and single-photon emission computed tomography. (d) Plot of similarity metric ρ_c as a function of blurring with the multiple curves representing different amounts of registration perturbations (the curve with x’s correspond to the initial registration). The other similarity metrics followed a similar trend. Abbreviation: PET = positron emission tomography.

Fig. 3.

(a) ⁹⁰Y PET/CT, ⁹⁰Y PET_+10mm/CT, and the ^99mTc macroaggregated albumin SPECT/CT for a patient that showed high values and a patient that showed intermediate values of similarity metric ρ_c. The similarity metric is plotted as a function of blurring for (b) the patient with high ρ_c and (c) the patient with intermediate ρ_c. The multiple curves correspond to different levels of registration perturbations, and the curve with x’s represents the initial registration. Abbreviations: CT = computed tomography; PET = positron emission tomography; SPECT = single photon emission computed tomography.

Fig. 4.

For each patient, the best value of (a) Lin’s concordance correlation coefficient, (b) Pearson’s correlation, (c) Spearman’s rank correlation coefficient, and (d) mean squared difference are plotted as a black triangle. The phantom-based (red dotted line) and patient-specific (a blue circle for each patient) acceptance criteria are also plotted. The x axis is the patient number. Number of patients meeting acceptance criteria is stated in each plot.

Fig. 5.

Dice similarity coefficient comparing SPECT and PET_10mm for contours corresponding to 3 threshold values. Each circle represents a patient data point. The dashed lines represent the smallest Dice similarity coefficient value measured from the 5 phantom acquisitions. Abbreviations: PET = positron emission tomography; SPECT = single photon emission computed tomography.