Combined Use of 123i-fp-cit Spect and 123i-mibg Scintigraphy for Differentiation of Progressive Supranuclear Palsy Subtypes and Parkinson's Disease

Background: This study was undertaken to investigate the utility of 123 I-ioupane ( 123 I-FP-CIT) single photon emission computed tomography (SPECT), 123 I-metaiodobenzylguanidine ( 123 I-MIBG) scintigraphy and both of these to differentiate among progressive supranuclear palsy (PSP), including typical cases and other subtypes, and Parkinson’s disease (PD). Methods: Twenty-ve patients with typical PSP (Richardson's syndrome; PSP-RS), 14 atypical ones (PSP-variants; PSP-V) and 42 PD who underwent both 23 I-FP-CIT SPECT and 123 I-MIBG scintigraphy within short intervals were enrolled. Speci ﬁ c binding ratio (SBR) of the striatum and midbrain and anteroposterior and asymmetry ratio of the striatal SBR on 123 I-FP-CIT SPECT and heart-to-mediastinum (H/M) ratio and washout rate (WR) on 123 I-MIBG scintigraphy were used as quantitative measures. The classier performance based on adaptive boosting was evaluated using ve-fold cross-validation for these measures.

important for deciding on treatments and management, accurate differentiation between PSP clinical subtypes other than the typical Richardson's syndrome and Parkinson's disease (PD) is often di cult because of the similarities in their neurological symptoms and lack of reliable disease-speci c biomarkers in the early stages [2,3,4]. Because appropriate clinical distinction among typical PSP, other PSP subtypes and PD has implications for prognosis, disease progression, and caregiver burden, it is valuable to attempt a differentiation among typical PSP, other PSP subtypes and PD based on objective methods other than the neurological symptoms and ndings alone [5].
The lack of speci c structural imaging ndings such as midbrain atrophy in the eld of imaging diagnosis, especially in cases with early stage PSP, highlights the need for sensitive functional imaging techniques to improve the diagnostic accuracy. Indeed, dopamine transporter (DAT) and serotonin transporter (SERT) imaging on single photon emission computed tomography (SPECT) is already being performed to differentiate PSP from PD [6,7,8,9,10]. These techniques delineate different aspects of the neurodegeneration and add some value to diagnose PSP including its subtypes [11,12]. However, presynaptic nigrostriatal degeneration is not a speci c pathological change. Therefore, DAT imaging is of limited value in differentiating between PSP and PD [13]. On the other hand, more speci c techniques including 123 I-metaiodobenzylguanidine (MIBG) scintigraphy, which can detect degeneration of the cardiac sympathetic system, are necessary to differentiate PSP from PD [14]. Similarly, the usefulness of the combination of DAT and cardiac sympathetic imaging has been reported in patients with parkinsonian syndrome [15,16]. Nevertheless, the utility of combination of DAT and cardiac sympathetic imaging has not been evaluated in PSP subtypes and PD patients. Considering the inhomogeneous and misleading neurological symptoms associated with the diverse pathological tau burden and distribution in the substantia nigra in PSP, it is plausible to hypothesize that the combination of DAT, SERT and cardiac sympathetic imaging would be useful in differentiating typical PSP, other clinical subtypes and PD. The aim of the present study was to investigate whether the combination of DAT, SERT and cardiac sympathetic imaging using 123 I-io upane ( 123 I-FP-CIT) SPECT and 123 I-MIBG scintigraphy can offer useful clues to differentiate among PSP, including typical cases and other subtypes, and PD.

Subjects
This was a retrospective study that evaluated the diagnostic value of 123 I-FP-CIT SPECT for differentiating PSP, including stereotype and subtypes, and PD patients. This study used data obtained at a single medical center, and was approved by our Institutional Review Board (IRB). The retrospective nature of this study made it di cult to acquire informed consent from all patients. Therefore, with the permission of IRB, we provided an opportunity for these patients to opt out. The privacy of all patients was completely protected. Patient backgrounds were standardized by applying the following inclusion criteria: (1) diagnoses of PSP and PD at the Department of Neurology and (2) the acquisition of both 123 I-FP-CIT SPECT and 123 I-MIBG scintigraphy within a short interval (less than two weeks for most patients). The diagnosis of PSP including typical cases (Richardson's syndrome; PSP-RS) and other subtypes (PSP-variants; PSP-V) was based on the current diagnostic criteria [1,17,18,19]. PD patients were diagnosed on the basis of the criteria set by the United Kingdom Parkinson's Disease Society Brain Bank [20]. Exclusion criteria were insu cient quality of the 123 I-FP-CIT SPECT images and failure of image analyses due to the presence of signi cantly abnormal ndings (e.g., large cerebral infarctions or hemorrhages). All patients were evaluated clinically, tested for L-dopa responses, and examined with SPECT between April 2014 and March 2017. Thirty-nine PSP (mean age, 73 ± 7 years; 28 men and 11 women) and 42 PD (mean age, 75 ± 7 years; 19 men and 23 women) patients were enrolled. 123 I-FP-CIT SPECT imaging and volume of interest analysis SPECT imaging was performed four hours after an intravenous injection of 167 MBq 123 I-FP-CIT (DaTSCAN®; Nihon Medi-Physics, Tokyo, Japan) on a dual-head gamma camera system PRISM-AXIS (SHIMADZU. Co., Kyoto, Japan), equipped with low-energy, high-resolution, and parallel-hole collimators. Energy windows were set at 140 keV ± 20%, and 90 views were obtained throughout 360 degrees of rotation (128 × 128 matrix, 3.0 mm/pixel). Data processing was performed on an ODYSSEY FX PRISM-AXIS (SHIMADZU. Co., Kyoto, Japan). All images were reconstructed using OSEM (iteration 3, subset 15) and then 3D smoothed with a Butterworth lter (order 8, cut-off 0.3 cycles/pixel). The brain volume was resliced after reorienting the axial planes along the anterior-posterior commissural line, and axial sections of 3.0 mm in thickness were produced. images of all PSP and PD patients were spatially normalized onto this in-house-made 123 I-FP-CIT SPECT template, as described previously, in the Montreal Neurological Institute (MNI) space [21]. The nal image format was 16-bit, with a size of 79 × 95 × 68 and voxel size of 2 × 2 × 2 mm.
To evaluate the striatal binding, bilateral striatal VOIs were created using an in-house-made 123 I-FP-CIT SPECT template and MRIcron (https://www.nitrc.org/projects/mricron). Considering previous studies that reported posterior dominant binding reduction in PD and uniform binding reduction in PSP, bilateral striatal VOIs were divided into anterior and posterior VOIs [22]. According to studies that reported lower SERT midbrain binding in PSP patients, the midbrain VOI was obtained with WFU PickAtlas tool in SPM8 [8,9,23]. Additionally, the bilateral occipital lobes were selected as a reference region. After spatial normalization and VOI placement, the presence of misregistration was visually evaluated. Speci c-tononspeci c binding ratios (SBR) in the targeting region (i.e., striatum and midbrain) were de ned as (mean counts of targeting region -mean counts of occipital lobe)/(mean counts of occipital lobe) [24]. Additionally, the anterior-to-posterior striatal ratio (StAP), which represented anterior striatal counts/posterior striatal counts, and asymmetry striatal ratio (StAS), which represented the less affected side striatal counts/more affected side striatal counts, were also calculated to evaluate the anteroposterior and asymmetry indices of the striatum according to the previously reported method.

I-MIBG myocardial scintigraphy and region of interest analysis
Early and delayed static images with a 128 × 128 matrix were obtained at 20 minutes and 4 hours after intravenous injection of 111 MBq of 123 I-MIBG, respectively. Planar images were obtained using the same dual-head gamma camera system, equipped with low-energy, general-purpose, and parallel-hole collimators. Regions of interest (ROI) were drawn manually around the whole heart and mediastinum. According to the standard method described previously, the heart-to-mediastinum (H/M) ratio was calculated from the average counts per pixel in the heart and mediastinum [25]. The washout rate (WR), which is an index of the rate at which MIBG is washed out between the early and delayed images, was also calculated according to the following formula: WR = [(early heart counts -early mediastinal counts) -(delayed heart counts -delayed mediastinal counts)]/(early heart counts -early mediastinal counts) x 100. For the comparison study, early and delayed H/M ratios and WR were used for analysis.

Machine Learning
All machine learning processes and statistical analyses were performed in the Python environment (version 3.7.3) by using the scikit-learn (version 0.19.1) and SciPy (version 1.2.1) [26]. A classi er algorithm based on adaptive boosting (called AdaBoost) was trained to evaluate the ability to discriminate between each type of pathological condition [27]. The AdaBoost classi er is a decision stump-based ensemble classi er that has advantages of algorithm transparency, robust discrimination ability, and a built-in feature selection process. The algorithm adaptively seeks the optimal combination of weak classi ers and weights assigned to each weak classi er. During each iteration, the algorithm selects weak classi ers to minimize the training error (sum of the misclassi ed sample weights) and down-weights correctly classi ed samples. Subsequently, the algorithm selects weak classi ers sensitive to misclassi ed samples during previous iterations based on sample weights until the performance of the classi er reaches a plateau. Based on the result of our experimental studies, the parameter of the classi er was used as n_estimators = 100.

Classi cation Performance Evaluation
A patient-strati ed ve-fold cross-validation scheme was repeated 100 times to mitigate bias towards a particular class and selection and classi er training bias. Predictive performance of the classi er and stability was evaluated by area under the receiver operator curve (AUC) and 95% con dence interval (CI), respectively. AUCs were compared with Mann-Whitney/Wilcoxon U statistics with Bonferroni adjustments.

Statistical analysis
Statistical analyses were performed using IBM SPSS statistics 21 (IBM SPSS Inc, Chicago, IL). The Kruskal-Wallis and Mann-Whitney U test for non-normally distributed data (age, gender, disease duration, Hohen-Yahr stage and H/M ratio) and one-way analysis of variance and unpaired t-test for normally distributed data (SBR) were performed for comparisons among patient groups. Comparisons of SBR among PSP-RS, PSP-V and PD patients were also performed. When a signi cant level was found in multiple comparisons, the Mann-Whitney U test or unpaired t-test was also performed. After Bonferroni corrections for multiple comparisons, differences were considered signi cant when p < 0.016.

Patient backgrounds
Patient characteristics are summarized in Table 1

VOI and ROI analyses results
The results of the VOI and ROI analyses are shown in Table 2

Machine learning results
For each pair-wise comparison, three different models were constructed and applied: FP-CIT model using 123 I-FP-CIT imaging metrics only; MIBG model using 123 I-MIBG metrics only; and FP-CIT-MIBG model using both 123 I-FP-CIT and 123 I-MIBG metrics. The diagnostic performance is detailed in Table 3  Feature metrics were ranked using the best performance classi ers for each pair-wise discrimination ( Table 4). These features included 4 FP-CIT features for the comparison between PSP-RS and PSP-V, 4 FP-CIT and 3 MIBG features for the comparison between PD and PSP-RS, and 3 MIBG features for the comparison between PD and PSP-V. Feature importance was normalized with respect to the best predictor.

Discussion
This is the rst study to evaluate the utility of the combined use of 123 I-FP-CIT SPECT and 123 I-MIBG scintigraphy to differentiate among PSP including various subtypes other than typical Richardson's syndrome and PD. We found that the combination of 123 I-FP-CIT SPECT and 123 I-MIBG scintigraphy could offer a valuable clue to differentiate PSP including both PSP-RS and PSP-V from PD. On the other hand, even with the use of VOI analyses results including statistically lower midbrain SBR and striatal anteroposterior ratio, 123 I-FP-CIT SPECT exhibited a low discrimination power (FP-CIT model with AUC of 0.530) to differentiate between PSP-RS and PD. Similarly, the discrimination power to differentiate PSP-V from PD was also insu cient (FP-CIT model with AUC of 0.570). These results recon rm the di culty in reliably distinguishing PD from other parkinsonian syndromes including PSP on the basis of the DAT imaging results alone [13].
Regarding PSP-RS, the results of the VOI analyses, namely a lower striatal anteroposterior ratio and midbrain SBR as compared with PD, were consistent with those of previous studies [6,8,9]. PSP-RS tends to have a more anterior dominant and profound low striatal SBR than that of PD [6]. The reduced midbrain SBR in PSP-RS is another well-known imaging nding [8,9]. These differences are surmised to re ect a more widespread decline of monoaminergic neurotransmitter systems due to PSP related neuropathologic changes in the brainstem areas containing SERT, DAT or noradrenergic transporter bearing neurons, such as the substantia nigra, locus ceruleus, and raphe nuclei [28,29,30].
On the other hand, compared with PSP-RS, PSP-V reveal relatively mild reduction of striatal and midbrain SBR and striatal anteroposterior ratio. To the best of our knowledge, no studies have evaluated the difference in the midbrain SBR between PSP-RS and PSP-V patients. Despite a lack of statistical difference, PSP-RS patients tended to have a lower midbrain SBR than PSP-V patients. Differences in the pathological severity have been recognized to affect both the clinical and radiological ndings in PSP patients [31]. Considering the more severe pathological changes of the brainstem noted in PSP-RS than in other subtypes, the assumption of a relatively lower midbrain SBR in PSP-RS patients is reasonable. In contrast to the two studies that reported lower caudate DAT activity impairment to be the key point to differentiate PSP subtypes including PSP-P and PSP-PGF from PD, no statistical differences were apparent between PSP-V and PD in our study [11,12]. This result would seem to be inconsistent with our VOI analyses. We assumed the major reason for this inconsistency to be the non-uniformity of the PSP subtypes including not only PSP-P and PSP-PGF but also PSP-CBS and PSP-C in this study.
Even with the use of machine learning, to improve the low discrimination power of DAT imaging, it is necessary to use more speci c functional imaging techniques to more accurately differentiate PSP from PD [32]. Some studies combining DAT and D2 receptor SPECT have shown that PD can be differentiated from PSP, whereas others have demonstrated otherwise [33,34,35]. Taking into account the diseasespeci city, a more promising technique would be cardiac sympathetic imaging with 123 I-MIBG scintigraphy, which can detect the myocardial adrenergic denervation caused by PD. A previous study combining DAT, D2 receptor and cardiac sympathetic imaging has reported a positive predictive value of 89%, and a negative predictive value of 97% to differentiate degenerative parkinsonism [36].
In this study, the importance of various features was suggested based on the results of the AdaBoost classi er algorithm. Delayed H/M ratio was found to be the most valuable of these in discriminating PSP from PD. Considering the degeneration of the cardiac sympathetic system in PD patients, it is reasonable to suggest that delayed H/M ratio on 123 I-MIBG scintigraphy is more useful than other indices on 123 I-FP-CIT SPECT to differentiate PSP including both PSP-RS and PSP-V from PD patients [37]. Meanwhile, midbrain SBR is of secondary usefulness in differentiating PSP-RS from PD. This result is also consistent with that of a previous study, in which a marked reduction of SERT in typical PSP patients compared to PD was con rmed [9]. For the differentiation between PSP-RS and PSP-V, striatal anteroposterior and asymmetry ratio and SBR of the striatum and midbrain on 123 I-FP-CIT SPECT have been shown to be of high normalized Gini importance. Previous studies evaluating PSP-V have reported a lower striatal uptake and higher asymmetric index in PSP-P and slightly higher caudate SBR in PSP-PGF than PSP-RS patients [11,12]. It has recently become apparent that the degree of neuropathological changes is inconsistent in PSP patients. Therefore, it is plausible to ascribe the different distribution of radiotracer uptake on 123 I-FP-CIT SPECT to differences in the degree of neurodegeneration of DAT and SERT in PSP-RS and PSP-V.
The relatively small number of patients, especially PSP clinical subtypes other than PSP-RS, is a limitation of the present study. Due to this, it was di cult to perform statistical analyses in each PSP clinical subtype. Therefore, PSP patients other than PSP-RS were grouped into a single entity as PSP-V. This may have affected the results of the VOI analyses and machine learning. Furthermore, our study may also have been limited by the absence of pathological diagnoses in most cases excluding three PSP-RS and one of the PSP-CBS patients. Considering these limitations, it will be necessary to investigate more cases with PSP clinical subtypes to further clarify the diagnostic value of the combined use of 123 I-FP-CIT SPECT and 123 I-MIBG scintigraphy for the differentiation among PSP-RS, PSP-V and PD patients.

Conclusion
Considering the di culty to differentiate PSP-V from PD and different clinical outcome between PSP-V and PSP-RS, it is important to make an accurate antemortem diagnosis. When it is di cult to differentiate patients with PSP-RS, PSP-V and PD, the combination of 123 I-FP-CIT SPECT and 123 I-MIBG scintigraphy, rather than either of these modalities alone, may be a useful differential diagnostic tool. Declarations acquisition. KS and IS contributed to the data analysis and interpretation. KS performed the statistical analysis. IA, KS and IS contributed to drafting of the manuscript together. The authors read and approved the nal manuscript.

Funding
No funding was received for this study.

Availability of data and materials
The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Ethics approval and consent to participate
We declare that all human and animal studies have been approved by the Institutional Review Board  Boxplots of 123I-FP-CIT VOI (a-d) and 123I-MIBG ROI (e-g) analyses results Striatal speci c binding ratio (SBR), striatal anteroposterior ratio, striatal asymmetry ratio and midbrain SBR on 123I-FP-CIT SPECT, and early and delayed heart-to-mediastinum ratio and washout rate on 123I-MIBG scintigraphy were compared among PSP-RS, PSP-V and PD patients. The box contains the median values (represented by a horizontal line within the box), 25th and 75th percentiles, and whiskers representing the minimum and maximum data.