Differential diagnosis of parkinsonian degenerative disorders in combination manual measurements with automated volumetry of the brain

Background : To compare brain morphological differences in progressive supranuclear palsy (PSP), multiple system atrophy with the parkinsonian variant (MSA-P), Parkinson’s disease (PD) and controls by manual and automated measurements and to explore the feasibility of these measurements in disease differentiation. Methods : Ninety-five PSP patients (48 males, mean age 67.9 y), 32 MSA-P patients (18 males, mean age 63.0 y), 136 PD patients (72 males, mean age 66.6 y) and 100 controls (50 males, mean age 66 y) were included. The 12 manual measurements were acquired. Relative brain structural volumes adjusted according to the intracranial volume (ICV) of different brain regions were also quantified. Differences among and between groups were evaluated. Receiver operating characteristic curve analysis was used to assess diagnostic performance and define cutoff values of these measures. Results: P/M area 2.0 displayed the highest diagnostic performance (AUC: 0.801) for distinguishing PSP from MSA-P or PD (sensitivity 69.5%, specificity 82.1%). Furthermore, the combination of morphological features in manual parameters (P/M area 2.0, MRPI and M/P diameter) and volume atrophy in the midbrain improved the PSP discrimination (AUC: 0.870, sensitivity 76.8%, specificity 83.9%). The relative volume of the putamen can better differentiate MSA-P from PSP and PD (AUC: 0.844, sensitivity 81.3%, specificity 75.3%). Similarly, the ability to differentially diagnose MSA-P increased most significantly (AUC: 0.927, sensitivity 87.5%, specificity 87.9%) when combing volume atrophy in the putamen with the caudate and manual parameter (M/P diameter). Conclusion: Manual and automated MR variables can reveal atrophy features of the brain and be helpful in the differential diagnosis.


BACKGROUND
Progressive supranuclear palsy (PSP) and multiple system atrophy (MSA) are two of the main progressive atypical parkinsonism syndromes (APS) that are frequently misdiagnosed as Parkinson's disease (PD). Up to 24% of patients clinically diagnosed with PD were classified as APS post mortem [1]. Accurate diagnosis of these three degenerative parkinsonian disorders based on clinical features, especially at an early stage, is challenging due to overlapping clinical features. Diagnosis based on a complete medical history includes a timeline of symptoms and recognition of the important clinical signs. Diagnostic accuracy is substantially influenced by clinical experience, and even among movement disorder specialists, the clinical diagnosis may change over time due to emerging clinical signs [2]. However, the correct differential diagnosis of PD, PSP and MSA is important for patient counselling and clinical research purposes. Therefore, supplementary supporting examinations are required for accurate diagnosis.
Radiographic signs have been reported to be of importance for differential diagnosis.
Magnetic resonance imaging (MRI) has played a significant role in parkinsonian disorders, and some morphological features that are helpful in differential diagnosis can be revealed on conventional MRI, such as abnormal iron accumulation with the disappearance of the "swallow tail" sign in PD [3], midbrain atrophy with the "hummingbird" sign or the "morning glory flower" sign in PSP [4,5], and abnormalities in the putamen, middle cerebellar peduncles (MCP), and pons with the presence of the "putaminal rim" sign, the MCP sign and the "hot cross bun" sign in MSA [6,7]. However, these MRI-based sign assessments have limitations. The disappearance of the "swallow tail" sign is helpful in distinguishing PD from controls [3,8], but not from APS [9,10]. Additionally, the "swallow tail" sign is not always visible at 3T, the procedure of imaging acquisition is not yet standardized and the swallow tail loss is not always bilaterally visible [11]. The "hummingbird" and "morning glory" signs are highly specific to PSP, and the MCP and "hot cross bun" signs are indicative of MSA; however, the detection sensitivity of these signs is low [6]. Additionally, these visual evaluation criteria are highly dependent on image quality, median location selection and the evaluator's experience [12,13].
Compared with morphological evaluation, quantitative analyses more objectively differentiate PD, PSP and MSA. Manual measures of the midbrain and pons (or their ratio) and calculation of the MR parkinsonism index (MRPI) identified PSP with a sensitivity up to 100% in some early studies [14,15]. However, the results were not as ideal in the subsequent studies. The sensitivity and specificity can vary widely when using manual measures [16]. MRPI 2.0 was first reported as an updated index for MRPI, which included the measurement of the third ventricle width. It was reported that this new MR index showed higher sensitivity (100%) and similar specificity (94.3%) in differentiating between PSP-P and PD [17]. However, further literature verifying these results is still lacking. MRI-based brain automatic segmentation volumetry can also differentiate parkinsonian disorders. Different patterns of brain atrophy in PSP, MSA and PD have been characterized [18]. It has been reported that both cortical and subcortical volumetry in supra-and infra-tentorial regions can be helpful in differentiating PSP, MSA and PD [16,19]. As manual measures have unstable diagnostic performance according to the present literature and few studies have combined the manual and automatic volumetric tools for PSP, MSA-P and PD differentiation, the aim of this study was first to further confirm the diagnostic performance of manual brainstem measurements and to determine whether there are more suitable cutoff values for disease differentiation. Second, we evaluated and compared both the automated volumetric differences of specific brain regions and manual parameters, trying to develop and optimize an efficient combined radiological diagnostic marker for PSP, MSA-P and PD differentiation. Nighty-five PSP patients (48 males, mean age 67.9 y, mean disease duration 3.8 y), 32 MSA-P patients (18 males, mean age 63.0 y, mean disease duration 2.9 y), 136 PD patients (72 males, mean age 66.6 y, mean disease duration 4.8 y) and 100 controls (50 males, mean age 66 y) were included ( Table 1). Two of the authors (X.D. and W.H.) with more than 10 years of experience in movement disorders determined the clinical diagnosis of patients together according to the consensus criteria [20][21][22]. The controls were selected from a community cohort, and the subjects all had a normal neurological examination and with no history of central nervous system diseases.

MRI acquisition
Brain MRI was performed with two 3.0 T units (Ingenia

Morphometric parameters based on manual measurements
The manual measurements including the third ventricle (3 rd V) and frontal horns (FH)  [14,15,17] by one neuroradiologist blinded to the clinical information. Interrater reliability studies were performed with images from ten randomly selected subjects assessed by another neuroradiologist.

Volumetric analysis based on automated segmentation
Relative brain volume measurements adjusted for the intracranial volume (ICV) of different brain regions were automatically quantified using AccuBrain® IV 1.0 software (BrainNow Medical Technology Limited, China) [23]. The brain regions included the midbrain, pons, cerebellum, putamen, pallidum, caudate, brain parenchyma and grey matter of different brain lobes ( Figure 2).

Statistical analysis
An intraclass correlation coefficient was used to assess the interrater agreement in the manual measurements. Differences across groups were compared by analysis of variance (ANOVA) followed by post-hoc least significant difference (LSD) test. P value <0.0083 was considered statistically significant with the Bonferroni correction for pairwise comparisons. Parameters with significant differences were further evaluated by receiver-operating characteristic (ROC) curves to compare the area under the curve (AUC). The optimal cut-off value balancing sensitivity and specificity was identified as the one corresponding to the maximum value of Youden's index, calculated as [sensitivity + specificity − 1]. The positive predictive value, negative predictive value and overall accuracy were also given. AUC values were interpreted as follows: 0.5 < AUC < 0.7 poor accuracy; 0.7 < AUC < 0.9 moderate accuracy; 0.9 < AUC < 1.0 high accuracy, AUC = 1 perfect accuracy. We also examined whether diagnostic accuracy was improved by using MR variables in combination and adjusting for patients' age and disease duration using logistic regression analysis with a forward stepwise procedure.
MR variables with with AUC ≥ 0.7 in univariate analyses were included one at a time.
The forward stepwise procedure stopped when additional variables did not contribute further significant information at the P ≤ 0.05 significance level. ROC curves were calculated for combinations of variables. Statistical analysis was performed with the Statistical Package for Social Science Software (SPSS, version 21.0, Chicago, IL) for Mac.

RESULTS
Tables 2 provides the results of the comparison of the quantitative MR parameters, including manual measurements and volumetric analysis, among MSA-P, PSP, PD patients and healthy controls. Tables 3 shows the pairwise comparison of the volumetric parameters across MSA-P, PSP, PD patients and healthy controls. Table 4 demonstrates the diagnostic performance of manual and volumetric parameters with AUC > 0.7 among only parkinsonian groups, including the cut-off values that provide an optimal balance of sensitivity and specificity, positive predictive value, negative predictive value and overall accuracy.

Interrater reliability
Ten participants were randomly selected from the main cohort for interrater reliability

Morphometric analysis of manual parameters
Significant differences among parkinsonian groups and healthy controls were found in the manual MR parameters at the group level ( Table 2). The midbrain diameter, the M/P diameter, midbrain area, and M/P area in the PSP were significantly smaller compared with those in all the others (P < 0.001 each). The PSP patients displayed the highest P/M area 2.0, MRPI and MRPI 2.0 values among all the groups (P < 0.001 each). The third ventricle was enlarged in the PSP with a larger 3 rd Vwidth (MSA P = 0.003; PD P < 0.001; HC P < 0.001). The 3 rd V-FH ratio of PSP was larger than that in the PD (P < 0.001) and healthy controls (P < 0.001), but not in the MSA-P group (P =0.019). The MSA-P patients had a smaller pons area compared with that in the other three groups (PSP P = 0.002; PD P < 0.001; HC P < 0.001), and a smaller pons diameter compared with that in the PSP (P = 0.001) and PD (P < 0.001).

Volumetric analysis
Significant difference in the volumetric comparison can be seen among groups in the supra-and infra-tentorial regions except the grey matter of the occipital lobe (P=0.334) ( Table 2). The midbrain volume in the PSP group was significantly distinguishable from that in the three other groups (MSA-P P = 0.001; PD P < 0.001; HC P < 0.001). The relative volume of the putamen was smallest in the MSA-P among the groups. The reduced volume in the pons, pallidum and cerebellum was observed in the MSA-P group compared with the PD patients and healthy controls. However, the difference between MSA-P and PSP was not statistically significant. The volume of the caudate was smaller in the MSA-P patients compared with the PSP (P < 0.001) and PD (P < 0.001) patients but not the healthy controls (P=0.389). The overall brain parenchyma and the grey matter of the frontal lobe were decreased in PSP patients compared with the other three groups. The reduced relative volume of the temporal and parietal lobe grey matter was found in the PSP and PD groups compared with the MSA-P group and healthy controls. The PD patients had a smaller volume of grey matter in the temporal and parietal lobes compared with the healthy controls. The healthy controls did not exhibit differences compared with the three Parkinson syndrome groups with respect to relative volume of the insular cortex. (Table 3.)

Morphometric analysis of manual parameters
The

Volumetric analysis
The relative volume of the midbrain showed the highest diagnostic performance in volumetric biomarkers for differentiating PSP from MSA-P and PD, with an AUC of 0.749 (95% CI: 0.687-0.812). Regarding differentiation of the MSA-P from the other two parkinsonian groups, the relative volume of the putamen became the most useful radiological biomarker among the other investigated ones, with an AUC > 0.8.

Combinations of manual and volumetric variables
We carried out a further analysis to determine whether an incremental accuracy could

DISCUSSION
Our study was a retrospective study with a large cohort applying MR variables of both manual measurements and automated volumetry to investigate the morphometric characteristics between PSP, MSA-P, and PD patients and healthy controls, and for differential diagnosis of parkinsonian disorders in clinical scenarios based on all these parameters. These measurements were based solely on conventional structural MRI and allowed the morphometric analysis of the brain quantitatively. Our study compared 12 manual radiographic biomarkers which included almost all manual measurements reported in the current literature. The measure method of these manual parameters was consistent with the previous literature and was highly reproducible between experts.
Subcortical volumetric changes in the basal ganglia (putamen, pallidum and caudate) were also discussed in the study.
In this study, the specific structural changes distinguishing PSP from the other two parkinsonian groups and healthy controls included 9 manual measures and 3 volumetric biomarkers. The midbrain atrophy in the PSP patients was significant, which was demonstrated not only in the manual measurements (midbrain diameter, midbrain area, M/P diameter, M/P area) but also in the volumetric analysis of the midbrain. These findings are consistent with the previous studies, which were in general agreement that midbrain measurements are smaller in PSP patients compared with MSA and PD patients. Other regional brain volumetric atrophy was also found in structures including the brain parenchyma and frontal lobe in PSP, which has been demonstrated by a number of group level studies [16]. A previous study showed that discernible frontal atrophy was present in only approximately 60% of PSP-RS patients by visual assessment [6].This further highlights the value of automated brain volumetry, which may identify cortical atrophy in the early phase of disease. Cordato et al. found a significant reduction of whole brain and frontal lobe volume in PSP compared to PD and healthy controls, but only frontal lobe volume contributed to the differential diagnosis [24]. Unfortunately, the atrophy of brain parenchyma and frontal lobe played a limited role in the differential diagnosis of PSP from MSA-P and PD, both of which showed a poor accuracy (AUC < 0.7) in this study. An enlarged third ventricle can be observed in the PSP patients through higher values of manual parameters in 3rdV width and 3rdV-FH ratio. The atrophy of the midbrain, superior cerebellar peduncles and an enlarged third ventricle were reflected in manual biomarkers, such as P/M area 2.0, MRPI, and MRPI 2.0, which represented various combinations of these structural changes.
We did not include the healthy controls in the analysis of the diagnostic performance of all different MR parameters. The inclusion of controls will make the AUC values better, but it obscures our ability to evaluate how well the measures actually differentiate parkinsonian syndromes. Thus, the results of the analyses without healthy controls are more likely to be up to the task of being applied to clinical diagnosis. Then, unfortunately, the widely accepted cutoff values for the midbrain diameter (9.35 mm), M/P diameter (0.52) and MRPI (13.55) did not yield perfect sensitivity (100%) and specificity (100%) in our study as has been reported in the literature [14,15]. Actually, the relatively moderate sensitivity and specificity levels for differentiating PSP have been previously reported [25][26][27]. A similar situation also occurred with the upgraded measurements (MRPI 2.0 and P/M area 2.0). The newly set cutoff values suitable for the patient cohort in this study differed from those in previous studies. This inconsistency may be more likely to reflect the application of these manual measurements in real clinical situations. However, despite the inconsistency, the manual parameters were still better suited to identify and differentiate (higher AUC) PSP from the other two parkinsonian groups compared with the automated volumetric analysis.
The modified biomarkers, including MRPI 2.0 and P/M area 2.0, all showed a better diagnostic performance in comparison with their original biomarkers (MRPI and M/P area) in differentiating PSP from MSA-P, PD. This finding was consistent with previous literature [17]. Among these measures, the P/M area 2.0 displayed the highest performance for PSP differentiation, and also had the highest accuracy. Furthermore, a combined indicator, the P/M area 2.0 in addition to the other 2 manual parameters (MRPI and M/P diameter) and volumetric quantitative MR markers (relative volume of the midbrain), helped increase diagnostic accuracy in discriminating PSP-RS from MSA-P, and PD to the greatest extent (from an AUC of 0.801 to 0.870). This result demonstrated that the comprehensive evaluation of these radiological markers for disease differentiation in the clinical settings is necessary.
Subcortical volume atrophy was found in the putamen and caudate in the MSA-P group compared with the PSP and PD groups. Volume atrophy in the putamen was most distinguishable and showed the best diagnostic performance in differentiating MSA-P from PSP and PD with moderate accuracy (AUC 0.844). Subcortical volume atrophy in the cerebellum, pons and pallidum were also identified in the MSA-P when compared with the PD and healthy controls. The higher degree of basal ganglia and cerebellar atrophy has been largely confirmed in the previous results and is in accordance with the evidence of MSA being a more aggressive and widespread disorder than PD [28]. The pons volumetric change can also explain the reduced values of the two manual biomarkers, including pons diameter and its area, but these two manual biomarkers did not result in good performance for MSA-P differentiation. Similarly, the combined marker, volumetric changes in the putamen and caudate in combination with a manual index (M/P diameter), improved the diagnostic accuracy to the greatest extent when discriminating MSA-P from PSP and PD (AUC>0.9). It has been reported that the volumes of the putamen and caudate were decreased in PD brains compared with control brains [29,30]. However, the volumetric difference of these two regions in PD was conflict with previous studies. The relative volume of the putamen and caudate was found to be lager in the PD patients compared with controls, which is opposite to the result in the previous literature. This contradictory result may indicate that the volume changes of the putamen in PD patients are not stable. Atrophy was also observed in cortical areas; including the temporal and parietal lobes, compared with the controls, which was in line with the findings reported in the early literature [31][32][33][34].
There were several limitations in this study. First, we used clinical criteria as an inclusion standard for PSP, MSA-P and PD patients, but pathological confirmation was not achieved in our patients. Thus, it is possible that in some patients, the clinical diagnosis could have been erroneous. However, it is difficult to select data for only pathologically confirmed patients. Furthermore, the diagnoses in this study were clinically evaluated by two of the authors (X.D. and W.H.) with more than 10 years of experience in movement disorders together according to the consensus criteria. Second, we did not specify clinical variants of PSP. This could have narrowed the differences found between the groups and therefore reduced the diagnostic performance of certain brain volumetric changes. However, in actual clinical scenarios, it is always more important and meaningful to differentiate the APS from PD than differentiate among the variants of APS. Additionally, this study discussed only a limited number of brain regions. Other brain structural changes that may be characteristic of certain diseases and helpful in disease differentiation were not included. A more comprehensive morphological imaging analysis is still needed in future studies.
In conclusion, the manual measurements and automated volumetry can identify different brain atrophy patterns among PSP, MSA-P, and PD patients and controls. The combination of radiological quantitative differences is helpful in the differential diagnosis of disease, although there is a current lack of agreement across studies on which specific regions should be used, and further validation of these results in independent cohorts is necessary. Whether disease-specific patterns of atrophy can be reliably detected in very early disease stages when the need for a correct diagnosis is most urgent remains to be proven.    The comparison of groups in brain parenchyma and the grey matter of different brain lobes. a. ANOVA followed by post-hoc least significant difference (LSD) test with Bonferroni-corrected significance level P = 0.05/6 = 0.0083.