Multi-parametric optimization of dopamine transporter SPECT software quantification for diagnosis of parkinsonian syndromes.

Objectives: Evaluate diagnostic accuracy of the parameters generated by a commercial quantification software for dopamine transporter (DaT) SPECT.

Methods: Diagnosis of a Parkinsonian syndrome (PS) was determined by a neurologist after at least two years of clinical follow-up and knowing result of DaT scan. Iodine-123 ioflupane DaT SPECT scans were quantified using DaTQUANT (GE Healthcare), which quantifies and compares regions of interest in the striatum, reporting Specific Binding Ratio (SBR) and Z-score for right and left sides. The ROI corresponding to the most abnormal side and therefore lowest quantification value (Z-score) was selected. We then fit a series of logistic regression models to predict PS status using the multiple variables provided by the quantification software. Both single- and multi-predictor models were explored where one predictor was fixed as the best predictor from the single-predictor models, using a maximum of two additional predictors. Prediction accuracy, sensitivity, and specificity were evaluated using leave-one-out cross validation. One-sided 95% confidence intervals for accuracy were constructed as bootstrap case cross-validation percentile intervals with bias reduction (1).

Results: A total of 129 DaTscans were included with 79 patients diagnosed with PS. Three single variables demonstrated the highest accuracy, 0.90, including whole putamen, anterior putamen, and posterior putamen. Using the same threshold, the anterior and posterior putamen also demonstrated the highest single parameter specificity of 0.88 while maintaining a sensitivity of 0.91. The entire putamen demonstrated a higher sensitivity, 0.94, but a lower specificity of 0.84. The threshold Z-scores were -1.9 for the posterior putamen, -1.4 for the anterior putamen, and -1.3 for the whole putamen. Based on our results and prior publications, the posterior putamen was selected for the construction of multi-parametric models (2-4). Six multi-parametric models produced the highest accuracy, 0.91. Three of these models demonstrated a higher sensitivity, 0.92, with a specificity of 0.88. Two models demonstrated a higher specificity, 0.94, but a lower sensitivity 0.89. The last model had sensitivity and specificity values of 0.91 and 0.90, respectively. The most sensitive model with the best accuracy as well as the narrowest confidence interval utilized the caudate, posterior putamen, and putamen asymmetry. The most specific model with the best accuracy as well as narrowest confidence interval utilized the posterior putamen, putamen asymmetry, and caudate asymmetry. The multi-parametric model with the highest overall sensitivity utilized the posterior putamen, caudate asymmetry, and the putamen to caudate ratio with accuracy, sensitivity, and specificity of 0.89, 0.96, and 0.78, respectively. CONCLUSION:The most accurate single variable predictors for a Parkinsonian syndrome are the anterior, posterior, and whole putamen with an accuracy of 0.90. A model incorporating the posterior putamen, caudate, and putamen asymmetry increases the accuracy to 0.91 with sensitivity of 0.92 and specificity of 0.88. A model incorporating the posterior putamen, putamen asymmetry, and caudate asymmetry attained the same accuracy (0.91) with higher specificity (0.94) but lower sensitivity (0.89). The multiparametric model with the highest sensitivity (0.96) included the posterior putamen, caudate asymmetry, and putamen to caudate ratio, but this significantly lowered the specificity (0.78).