Impact of the cerebrospinal fluid-mask algorithm on the diagnostic performance of ¹²³I-Ioflupane SPECT: an investigation of parkinsonian syndromes

This single-center retrospective study included 529 consecutive patients who underwent DAT SPECT from February 2014 to May 2017. Of these patients, 208 were excluded from the study on account of having duplicated or clinically undiagnosed cases (n = 14 and 177, respectively) or insufficient image quality (n = 17), which included cases with cerebral hemorrhage or brain infarction. Concerning the 177 patients who were clinically undiagnosed cases, they did not meet the diagnostic criteria for any disease finally. In addition, some patients who were difficult to follow due to hospital change were also excluded. Of the remaining 321 patients (median age, 68.9 years; range, 17–91 years; men/women, 176/145), 163 with PS and 158 without PS (NPS) were included in the assessment of the accuracy of the SBR index for diagnosing PS (analysis 1). Of the 163 patients with PS, 114 were diagnosed with PD based on the clinical diagnostic criteria of the UK Parkinson’s disease society brain bank [12]; the remaining 49 patients were diagnosed with atypical PS, including clinical DLB (n = 22), multiple system atrophy (MSA) (n = 11), and progressive supranuclear palsy (n = 16), on the basis of established diagnostic criteria [13‐15]. Of the 11 patients with MSA, eight were clinically diagnosed with MSA-Parkinsonism and three were diagnosed with MSA-Cerebellar. Patients with essential tremor, drug-induced Parkinsonism, and normal pressure hydrocephalus were included in the NPS group. In addition, the NPS group included patients whose symptoms, such as tremor, improved during follow-up and were not clinically diagnosed with a degenerative disease. Of the 321 patients, 275 who had CT or MRI images were included in the evaluation of the impact of ventricular dilatation on the CSF-mask algorithm (analysis 2). Based on the Evans index (EI) calculated with CT or MRI images, we divided the enrolled cases into three groups according to the extent of ventricular dilatation: low, middle, and high EI groups (n = 92, 92, and 91, respectively). A flow diagram presents the number of participants at each stage of the study (Fig. 1).

The institutional review board of Keio University School of Medicine granted permission for this retrospective review of imaging and clinical data and waived the requirement for obtaining informed consent from the patients (approval number: 20150441).

Using the Discovery NM/CT 670 or Discovery NM 630 (GE Healthcare, Milwaukee, WI) mounted with a FAN beam collimator, SPECT images were acquired 3 h after the injection of ¹²³I-Ioflupane (185 MBq). The imaging parameters were as follows: matrix size, 128 × 128; pixel size, 4.4 mm; slice thickness, 4.4 mm; and energy window, 159 keV ± 10%. The projection data acquired for 30 min were reconstructed on a Xeleris workstation (GE Healthcare). The ordered-subset expectation maximization method (iterations, three; subset, ten) and a Butterworth filter (critical frequency, 0.5; power, 10.0) were applied to the SPECT images. Both data with and without attenuation correction were generated (no correction [NC]; attenuation correction [AC]). Scatter correction was not used.

We used a commercially available software package for VOI based analysis: DaTView (AZE Co., Ltd., Tokyo, Japan). This software enables the semi-automatic calculation of the SBR index based on the Southampton method and mounts a function of the CSF-mask algorithm.

The CSF-mask algorithm is a threshold process that eliminates low counts caused by brain atrophy or ventricular dilatation. In the CSF-mask algorithm, the threshold is defined as follows: where \( \overset{\sim }{x} \) and σ are the median and standard deviation, respectively, of the background histogram with a Gaussian fit normal distribution and k is a coefficient. Previous reports informed our use of 1.0 as the k value [8, 10].

The SBR with the CSF-mask algorithm is calculated according to Southampton method as follows:

Where V_s is the striatal volume (assumed to be 11.2 ml); Ctc_VOI represents the total counts in the striatal VOI, excluding the threshold value; C_cr is the count concentration in the reference background, excluding the threshold value; and Vc_VOI is the volume of the striatal VOI, excluding the threshold volume.

We calculated the SBR indices with and without CSF-mask correction according to the method described above. Figure 2 shows the VOI template as well as the CSF-mask-algorithm-corrected image without the CSF-low-count areas (regions encompassed by the red line).

We assessed ventricular size using the EI, which is usually used to diagnose normal pressure hydrocephalus [16]. The EI is defined as the maximal width of the frontal horns of the lateral ventricles divided by the maximal internal diameter of the skull at the same level in axial MR or CT images. In this study, the EI was measured manually using CT (n = 119) or MR (n = 156) images obtained within 6 months of the DAT SPECT examination. The calculated EIs informed the delimitation of the enrolled cases into three approximately equally sized groups: low EI, middle EI, and high EI.

The Fisher’s exact test or t test was used for comparisons of age and sex distribution. The Mann-Whitney U test was used to compare the SBR indices between the PS and NPS groups. A receiver-operating characteristics (ROC) analysis was performed to evaluate the area under the curve (AUC). The DeLong method was used to examine the difference between the two AUCs [17]. The sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and accuracy of the SBR indices were calculated using the optimal cut-off values determined based on the ROC curves. Differences with p values of < 0.05 (two-sided) were considered to be statistically significant.

The Fisher’s exact test, the t test, and the Mann-Whitney U test were performed using SPSS software (version 25; SPSS Inc., Chicago, IL). The ROC analysis was performed using the statistical package R (version 3.2.2; available as a free download from http://​www.​r-project.​org).

Table 1 shows the characteristics of the included patients for analysis 1. No significant differences were observed with respect to age or the sex ratio between the PS and NPS groups. Figure 3 shows the box-and-whisker plots of the SBRs with and without the CSF-mask algorithm. The mean SBRs with and without the CSF-mask algorithm of patients with NPS were significantly higher than that of patients with PS (4.72 ± 1.18 vs. 2.62 ± 1.02, p < 0.001, Fig. 3a; 5.54 ± 1.34 vs. 3.32 ± 1.18, p < 0.001, Fig. 3b; 5.09 ± 0.97 vs. 3.10 ± 0.96, p < 0.001, Fig. 3c; 5.72 ± 1.13 vs. 3.68 ± 1.11, p < 0.001, Fig. 3d; respectively).

Figure 4 shows the results of the ROC analyses. The diagnostic performance of the SBR with the CSF-mask algorithm was significantly higher than that of the SBR without the CSF-mask algorithm (NC, AUC = 0.917 [95% confidence interval (CI) 0.887–0.947] vs. 0.895 [95% CI 0.861–0.929], p < 0.001, Fig. 4a; AC, AUC = 0.930 [95% CI 0.902–0.957] vs. 0.903 [95% CI 0.870–0.936], p < 0.001, Fig. 4b).

Table 2 provides a summary of the sensitivity, specificity, PPV, NPV, and accuracy of the SBRs with and without the CSF-mask algorithm. The cut-off values for the SBRs with and without the CSF-mask algorithm were 3.80 and 4.58 in NC, and 3.97 and 4.78 in AC, respectively.

Table 3 shows the characteristics of the included patients for analysis 2. The mean EIs of the low, middle, and high EI groups were 0.248 (0.215–0.265, n = 92), 0.279 (0.265–0.295, n = 92), and 0.325 (0.296–0.431, n = 91), respectively. Figure 5 presents representative cases of each EI group. Figure 6 presents the results of the ROC analyses. When not corrected for attenuation, in the low EI group, no significant difference was observed between the AUC of the CSF-mask and that of the non-CSF-mask (AUC, 0.942 [95% CI 0.898–0.986] vs. 0.927 [95% CI 0.877–0.978], respectively; p = 0.11; Fig. 6a). In the middle and high EI groups, the CSF-mask performed better than the non-CSF-mask (middle EI group, AUC = 0.894 [95% CI 0.825–0.963] vs. 0.872 [95% CI 0.798–0.947], p < 0.05, Fig. 6b; high EI group, AUC = 0.931 [95% CI 0.883–0.978] vs. 0.900 [95% CI 0.840–0.961], p < 0.01, Fig. 6c). When corrected for attenuation, significant differences in the AUC were observed in all three EI groups (low EI, AUC = 0.961 [95% CI 0.924–0.998] vs. 0.942 [95% CI 0.895–0.988], p < 0.05, Fig. 6d; middle EI, AUC = 0.905 [95% CI 0.843–0.968] vs. 0.872 [95% CI 0.800–0.944], p < 0.005, Fig. 6e; high EI, AUC = 0.954 [95% CI 0.917–0.991] vs. 0.917 [95% CI 0.862–0.973], p < 0.005, Fig. 6f).