A New MRI Measure to Early Differentiate Progressive Supranuclear Palsy From De Novo Parkinson’s Disease in Clinical Practice: An International Study

Abstract

Background:

Enlargement of the third ventricle has been reported in atypical parkinsonism. We investigated whether the measurement of third ventricle width could distinguish Parkinson’s disease (PD) from progressive supranuclear palsy (PSP).

Methods:

We assessed a new MR T1-weighted measurement (third ventricle width/internal skull diameter) in a training cohort of 268 participants (98 PD, 73 PSP, 98 controls from our center) and in a testing cohort of 291 participants (82 de novo PD patients and 133 controls from the Parkinson’s Progression Markers Initiative, 76 early-stage PSP from an international research group). PD diagnosis was confirmed after a 4-year follow-up. Diagnostic performance of the third ventricle/internal skull diameter was assessed using receiver operating characteristic curve with bootstrapping; the area under the curve of the training cohort was compared with the area under the curve of the testing cohort using the De Long test.

Results:

In both cohorts, third ventricle/internal skull diameter values did not differ between PD and controls but were significantly lower in PD than in PSP patients (P < 0.0001). In PD, third ventricle/internal skull diameter values did not change significantly between baseline and follow-up evaluation. Receiver operating characteristic analysis accurately differentiated PD from PSP in the training cohort (area under the curve, 0.94; 95% CI, 91.1–97.6; cutoff, 5.72) and in the testing cohort (area under the curve, 0.91; 95% CI, 87.0–97.0; cutoff,: 5.88), validating the generalizability of the results.

Conclusion:

Our study provides a new reliable and validated MRI measurement for the early differentiation of PD and PSP. The simplicity and generalizability of this biomarker make it suitable for routine clinical practice and for selection of patients in clinical trials worldwide.

Differentiating Parkinson’s disease (PD) from other neurodegenerative diseases such as progressive supranuclear palsy (PSP) can be difficult in the first years of the disease, when the symptoms are few and mild and the response to dopaminergic treatment is less defined, but is important for the patient’s treatment and prognosis.^1–3

Several quantitative MRI measurements have proven to be useful biomarkers in differentiating between PD and PSP patients.^4,5 A recent review from the Movement Disorder Society–endorsed Study Group⁵ reported that the Magnetic Resonance Parkinsonism Index (MRPI) is one of the most reliable imaging biomarkers for PSP, supporting clinical diagnoses in these patients. Several studies have also demonstrated that enlargement of the third ventricle (3^rdV) commonly occurs in atypical parkinsonism, above all PSP, compared with PD patients and control subjects.^6–10 Recently, a new version of MRPI including third ventricle width measurement, showed excellent diagnostic performance in distinguishing different PSP phenotypes from PD⁶ and in predicting the appearance of a PSP phenotype in patients with an initial diagnosis of PD.³ MRPI and MRPI 2.0, however, require specific image reconstruction and expertise and are suited to research rather than routine clinical practice.

In the current study, we developed and validated a new simple measurement of the 3^rdV width/internal skull diameter ratio (3^rdV/ID) in 2 large independent cohorts (training and testing) for the differentiation of PSP from PD patients in the early and late stages of the diseases.

Materials and Methods

Patients

The training cohort included 269 participants (98 treated PD, 73 late-stage PSP, 98 controls) from our Movement Disorder Centre of Magna Graecia University, Catanzaro, Italy, and the testing cohort included 291 participants (82 de novo PD and 133 controls from the Parkinson’s Progression Markers Initiative (PPMI)^11,12 and 76 early-stage PSP patients from an international research group¹³). Patients and controls in the training cohort were consecutively recruited between April 2010 and July 2019 (PD patients and controls, 2010–2015; PSP patients, 2010–2019). Clinical diagnosis of PD and PSP was made by a physician with more than 10 years of experience in movement disorders, according to international diagnostic criteria.^14,15 PSP patients enrolled before 2017 were diagnosed according to previous diagnostic criteria and expert guidelines^16,17 and were reclassified according to recent diagnostic criteria in PSP-Richardson’s syndrome (PSP-RS) and PSP-parkinsonism (PSP-P).¹⁵ For each patient, a complete medical history and neurological examination were performed, including the Unified Parkinson’s Disease Rating Scale part III (UPDRS-III)¹⁸ in the off-state (off medications overnight), the Hoehn and Yahr (H-Y) rating scale,¹⁹ and the Mini–Mental State Examination (MMSE). Levodopa response was assessed both in the off-state and 2 hours after drug administration, and was considered positive if a decrease of ≥30% on the UPDRS-III score was observed. PD patients were followed clinically every year for 4 years, and all patients repeated the MRI examination at the end of the follow-up. The initial PD diagnosis was confirmed at the end of a 4-year follow-up evaluation in all patients. A subgroup of 60 PSP, 85 PD, and 64 control subjects has already been reported in recent studies.^3,6

Eighty-two de novo PD patients (untreated patients with PD diagnosis for 2 years or less at screening) and 133 controls from the PPMI database were included in the testing cohort.^11,12 To select patients from the PPMI database, we considered all PD patients with disease onset ≥ 50 years in whom the PD diagnosis was confirmed after a 4-year follow-up, with available 3-dimensional (3-D) T1-weighted MRI images obtained both at baseline and at follow-up evaluations. We considered demographic and clinical data, including sex, age at examination, age at disease onset, disease duration, UPDRS-III score in the off-state, H-Y score, Montreal Cognitive Assessment (MoCA) score,²⁰ and levodopa response. The 76 PSP patients (59 PSP-RS and 17 PSP-P) included in the testing set belonged to an international early-stage PSP cohort reported in a recent study.¹³ Most of these patients were followed up for at least 24 months. The demographic and clinical data available included sex, age at examination, age at disease onset, disease duration, UPDRS-III score in the off-state, H-Y score, and MMSE score.

Exclusion criteria for our PD and PSP patients consisted of a history of neuroleptic use within the previous 6 months, clinical features suggestive of other diseases, normal striatal uptake on 123I-FP-CIT-single-photon emission computed tomography, and MRI abnormalities such as lacunar infarctions in the basal ganglia and/or subcortical vascular lesions with diffuse periventricular signal alterations. We enrolled PD patients with disease onset >50 years to exclude most forms of genetic PD that typically occur as early onset.²¹ None of the control participants had a history of neurologic, psychiatric, or other major medical illnesses. Full inclusion and exclusion criteria can be found online for PPMI participants (www.ppmi-info.org) and elsewhere for the international PSP cohort.¹³ We also excluded subjects from the study who showed Evans Index (EI) > 0.32 associated with callosal angle (CA) < 100°, a combination of MRI biomarkers strongly suggestive of normal pressure hydrocephalus.²²

All study procedures and ethical aspects were approved by the institutional review board (Magna Graecia University review board, Catanzaro, Italy). Each PPMI recruitment site received approval from an institutional review board or ethics committee on human experimentation before PPMI study initiation. Written informed consent for research was obtained from all individuals participating in the study.

MR Imaging Protocol

All patients and controls in the training cohort underwent brain MRIs with a 3T MR750 General Electric scanner and an 8-channel head coil, with a recently described MRI protocol.⁶

Patients and controls from the international cohorts underwent brain MRIs with 3T (65 PSP, 82 PD, and 92 controls) or 1.5-T scanners (11 PSP, 41 controls), with a protocol including a T1-weighted volumetric image. Further details can be found in the MRI technical operation manual at ppmi-info.org for PPMI patients and controls, and elsewhere for PSP patients.¹³

MRI Measurements

The 3^rdV width was normalized dividing by the largest internal skull diameter (ID), and the ratio was multiplied by 100. Both measurements were manually performed, and the 3^rdV/ID ratio was calculated in PD patients both at baseline and at 4-year follow-up, in PSP patients, and in controls. Using 3-D T1-weighted images at the midsagittal plane, the subcallosal line²³ was used to generate an axial slice for measuring the third ventricle width at the level of its maximum dilatation, as the linear distance between the 3^rdV lateral borders in the central portion of the ventricle. The maximum ID was also measured on the same axial slice (Fig. 1). The 3^rdV width and the ID measurements were performed by 1 trained rater (M.G.B.) who was blinded to clinical diagnosis. A second trained rater (R.V.), also blinded to clinical diagnosis, independently performed the measurements in 60 participants (20 PSP, 20 PD, 20 healthy controls), and the interrater reliability was calculated to verify the agreement in manual measurements between the 2 raters. One of the 2 raters (M.G. B.) also performed a second evaluation 1 month after the first in the same 60 participants to assess the intrarater reliability. EI and CA were manually measured by 1 rater (M.G.B.) in all participants as previously described.²²

FIG. 1.

Subcallosal axial T1-weighted volumetric MR images showing the measurement of the third ventricle width and the internal skull diameter (ID) in (A) a patient with progressive supranuclear palsy (PSP), (B) a patient with Parkinson’s disease (PD), and (C) a control subject. (D) Axial slices generated using the subcallosal line on 3-D T1-weighted sagittal images. Measurements were performed at the level of the third ventricle’s maximum dilatation as the largest left-to-right width between the lateral borders of the ventricle in its central portion. The maximum ID was also measured on the same axial slice. The third ventricle width was normalized dividing by the ID, and the ratio value was multiplied by 100. Images show marked dilatation of the third ventricle in the PSP patient compared with the PD patient and the control subject.

Statistical Analysis

The 3^rdV width, ID, and 3^rdV/ID ratio values were compared using analysis of covariance in generalized linear models, with age, sex, and disease duration as covariates, followed by the Tukey test; MMSE and scanner field strength (1.5 and 3.0 T, respectively) were also analyzed as covariates in the training and the testing cohorts, respectively. Logistic regression models using 3^rdV/ID and age as independent factors and diagnosis as dependent variables were used to calculate the odds ratio (OR) for PSP in both cohorts. As no collinearity was found and age did not contribute significantly to the prediction of diagnosis (P = 0.87 in the training cohort and P = 0.66 in the testing cohort), the regression model was performed using 3^rdV/ID as a unique predictor. Logistic regression models in the training and testing cohorts were compared using Chow’s test. We assessed sensitivity, specificity, positive predictive value (PPV), negative predictive value, diagnostic accuracy, and area under the curve (AUC) of 3^rdV/ID in differentiating PSP from PD patients in both the training and testing cohorts. Optimal cutoff levels and 95% confidence intervals were calculated using the pROC software package with bootstrapping (n = 10,000 iterations).²⁴ The diagnostic performance of this biomarker in the 2 cohorts was compared using the De Long test. The cutoff values obtained in each cohort were then tested in the other independent cohort, and the diagnostic performance obtained using the 2 cutoff values was compared using McNemar’s test in each cohort to assess the generalizability of the findings. Further details are provided in the supplementary materials.

Results

Patients

The demographic, clinical, and imaging data of PSP patients, PD patients, and control subjects in the training and the testing cohorts are summarized in Table 1 and Table S1.

TABLE 1.

Demographic, clinical, and imaging data of patients with progressive supranuclear palsy, Parkinson’s disease, and control subjects in the training and testing cohorts

	Training cohort				Testing cohort
Data	PSP (n = 73)	PD (n = 98)	Control subjects (n = 98)	P	PSP (n = 76)	PD (n = 82)	Control subjects (n = 133)	P
Sex (M/F)	43/30	61/37	50/48	0.268a	46/30	54/28	84/49	0.795a
Age at examination, yearsb	71.4 ± 6.6h,j	63.1 ± 7.7	63.4 ± 8.1	< 0.001c	69.7 ± 6.2h,j	62.8 ± 7.8	63.9 ± 7.7	< 0.001c
Age at disease onset, yearsb	65.0 ± 6.4j	59.5 ± 7.9	/	< 0.001d	67.4 ± 6.2j	62.3 ± 7.6	/	< 0.001d
Disease duration, yearsb	6.4 ± 2.9j	3.6 ± 3.5	/	< 0.001d	2.2 ± 0.8j	0.5 ± 0.7	/	< 0.001d
MMSE scoree	22 (6–29)h,j	27 (19–30)g	28 (24–30)	< 0.001d	25 (12–29)	/	/	/
MoCA scoree	/	/	/	/	/	28 (21–30)h	28 (26–30)	< 0.001d
UPDRS-III scoree	43 (20–76)h,j	30 (8–46)	/	< 0.001d	38 (10–67)j	20 (4–42)	/	< 0.001d
H-Y scoree	4 (2–5)	2 (1–3)	/	< 0.001d	3 (2–5)j	2 (1–2)	/	< 0.001d
Levodopa responsiveness, n (%)	3 (4.1)	80 (81.6)	/	< 0.001a	/	/	/	/
Brain MRI quantitative measurements
3rdV width, mmb	9.59 ± 2.46h,j	4.76 ± 1.93	4.20 ± 1.78	< 0.001f	9.54 ± 2.52h,j	5.20 ± 2.26	4.90 ± 2.04	< 0.001f
ID, mmb	132.5 ± 6.0	133.4 ± 5.7	132.0 ± 5.4	0.167f	134.2 ± 6.1i	135.9 ± 5.8	134.5 ± 6.2	0.082f
3rdV/ID ratiob	7.23 ± 1.79h,j	3.57 ± 1.43	3.17 ± 1.32	< 0.001f	7.10 ± 1.81j,h	3.80 ± 1.60	3.62 ± 1.44	< 0.001f

ANCOVA, analysis of covariance; PD, Parkinson’s disease; PSP, progressive supranuclear palsy; MMSE, Mini–Mental State Examination; MoCA, Montreal Cognitive Assessment; UPDRS-III, Unified Parkinson’s Disease Rating Scale part III; H-Y, Hoehn and Yahr rating scale; 3^rdV, third ventricle; ID, internal skull diameter.

^aFisher’s exact test.

^bData are expressed as mean ± standard deviation.

^cKruskal-Wallis test followed by pairwise Wilcoxon rank sum test.

^dANCOVA with age and sex as covariates, followed by Tukey test.

^eData are expressed as median (range).

^fANCOVA with age and sex as covariates, followed by the Tukey test; MMSE and scanner field strength (1.5 and 3.0 T, respectively) were also analyzed as covariates in the training and in the testing cohorts, respectively. Brain MRI quantitative measurements were also compared between PSP and PD patients with the ANCOVA test including disease duration as a covariate, with the following results: 3rdV width, P < 0.001 in both cohorts; ID, P > 0.05 (not significant) in both cohorts; 3rdV/ID, P < 0.001 in both cohorts. All P values were corrected according to Bonferroni.

^gPatients versus controls; P < 0.05.

^hPatients versus controls; P < 0.001.

ⁱPSP versus PD; P < 0.05.

^jPSP versus PD; P < 0.001.

The training cohort from our center included 73 late-stage PSP patients (44 PSP-RS and 29 PSP-P) with a mean disease duration of 6.4 ± 2.9 years, 98 treated PD patients with a mean disease duration of 3.6 ± 3.5 years at baseline, and 98 control subjects. The international testing cohort included 76 early-stage PSP patients (59 PSP–RS and 17 PSP-P) with a mean disease duration of 2.2 ± 0.8 years, 82 de novo PD patients with a mean disease duration of 0.5 ± 0.7 years, and 133 control subjects (Table 1). In both cohorts, PSP patients were significantly older and had longer disease duration than PD patients; thus, all analyses were corrected for age at examination and disease duration. In the training cohort PSP patients showed higher cognitive impairment than PD patients, and 3^rdV values were corrected for MMSE scores (Table 1). In the testing cohort, cognitive performances of PD and PSP patients were evaluated with MoCA and MMSE, respectively; thus, comparison of cognitive impairment between these 2 groups was not possible. (Table 1). In both cohorts, PSP patients showed higher disease severity than PD patients (Table 1). As shown in Table S1, PD and PSP patients in the testing cohort had a shorter disease duration and lower disease severity than those in the training cohort.

In both cohorts, PD patients showed higher disease severity at follow-up than at baseline, reflecting disease progression over the 4-year-follow-up period (Table S2). Our cohort also showed higher cognitive impairment at follow-up than at baseline, which was not found in the PPMI cohort, probably because of the shorter disease duration. Levodopa responsiveness was present in 100% of our PD patients and in 96% of the PPMI patients at follow-up evaluation (Table S2).

Morphometric Measurements

In both cohorts the 3^rdV/ID ratio values were higher in PSP patients than in PD patients and controls after correcting for age, sex, disease duration, MMSE, and scanner field strength, whereas no differences were found between PD patients and control subjects (Table 1). The 3^rdV/ID ratio values of PD patients, PSP patients, and controls in the training and testing cohorts are shown in Table S1. In PD patients, there were no significant differences in the 3^rdV/ID ratio after correcting for age between baseline and follow-up evaluations in both cohorts (Table S2). No difference in 3^rdV/ID values was found between the PSP-RS and PSP-P phenotypes (Table S3).

The odds ratio (OR) for PSP was 4.0 (95% CI, 2.8–6.5) and 3.1 (95% CI, 2.3–4.5) for each 1-unit increase in 3^rdV/ID in the training and testing cohorts respectively, thus demonstrating that the 3^rdV/ID values were strongly associated (P < 0.0001) with the probability of having PSP and that this probability increased with larger values. In contrast, the probability of having PD decreases with higher 3^rdV/ID values. There were no differences between the logistic regression models obtained in the training and testing cohorts (P = 1; Fig. 2).

FIG. 2.

Probability of having PSP or PD for each 3^rdV/ID value in the training (red) and testing (blue) cohorts obtained using logistic regression models. The probability of having PSP increased with higher 3^rdV/ID values, whereas the probability of having PD decreased with larger 3^rdV/ID values. There were no differences between the logistic regression models obtained in the training and testing cohorts (P = 1).

Diagnostic performance of the 3^rdV/ID in distinguishing between PSP and PD in both cohorts is shown in Table 2 and Figure 3. This MRI measurement showed a high AUC in the training cohort (AUC, 0.94; 95% CI, 0.91–0.98), which was confirmed in the international testing cohort (AUC, 0.91; 95% CI, 0.87–0.96). The De Long test showed no differences in the diagnostic performance of this biomarker between the training and the testing cohorts (D = 0.99, P = 0.320). To generalize the feasibility of our findings, the optimal 3^rdV/ID cutoff value obtained in each cohort was also tested in the other independent cohort. McNemar’s test showed no differences in the performance obtained with the 2 cutoffs in each cohort (P = 1 in the training chort and P = 0.37 in the testing cohort).

TABLE 2.

Diagnostic performance of the third ventricle width/internal skull diameter for the differentiation of patients with progressive supranuclear palsy from those with Parkinson’s disease, in the training and testing cohorts

Cutoff and statistical values	3rdV/ID
Late-stage PSP patients versus PD patients, training cohort
Cutoff value	≥5.72 (4.40–6.07)
Sensitivity (%)	84.9 (75.3–98.6)
Specificity (%)	92.9 (74.5–99.0)
PPV (%)	89.5 (74.0–99.2)
NPV (%)	89.4 (83.8–98.7)
Accuracy (%)	88.9 (83.6–93.0)
AUC (%)	94.3 (91.1–97.6)
Early-stage PSP patients vs. de novo PD patients, testing cohort
Cutoff value	≥5.88 (4.48–6.09)
Sensitivity (%)	81.6 (71.0–94.7)
Specificity (%)	93.9 (76.8–98.8)
PPV (%)	92.1 (78.9–98.4)
NPV (%)	84.5 (77.8–94.1)
Accuracy (%)	87.3 (82.3–91.8)
AUC (%)	91.5 (87.0–96.0)
Exporting the cutoff value obtained in the training cohort into the testing cohort
Cutoff value	≥5.72
Sensitivity (%)	80.3
Specificity (%)	87.8
PPV (%)	85.9
NPV (%)	82.8
Accuracy (%)	84.2
Exporting the cutoff value obtained in the testing cohort into the training cohort
Cutoff value	≥5.88
Sensitivity (%)	76.7
Specificity (%)	95.9
PPV (%)	93.3
NPV (%)	84.7
Accuracy (%)	87.7

PSP, progressive supranuclear palsy; PD, Parkinson’s disease; 3^rdV/ID, third ventricle width/internal skull diameter ratio; PPV, positive predictive value; NPV, negative predictive value; AUC, area under the curve.

FIG. 3.

Receiving operating characteristic (ROC) curves for the 3^rdV/ID in differentiating between PSP and PD patients, in the training cohort (red) and in the testing cohort (blue). The training cohort included 73 PSP patients and 98 PD patients; the testing cohort included 76 PSP patients and 82 PD patients.3^rdV/ID, third ventricle width/internal skull diameter; AUC, area under the curve.

All manual measurements showed excellent intrarater and interrater reliability (3^rdV width: intrarater ICC, 0.997; interrater ICC, 0.997; ID: intrarater ICC, 0.992; interrater ICC, 0.993; 3^rdV/ID ratio: intrarater ICC, 0.998; interrater ICC, 0.997).

Discussion

In this study in 2 large independent cohorts we developed and validated a new MRI measurement of the third ventricle width (3^rdV/ID ratio), which accurately differentiated PD from PSP patients in both the early and the late stages of the diseases.

There is evidence that the enlargement of the third ventricle is a common radiological feature in PSP, supporting the hypothesis that the 3^rdV width measurement may help to differentiate PSP from PD.^6–10 Several studies have investigated the third ventricle size in combination with substantia nigra hyperechogenicity in PSP and PD patients using transcranial sonography, showing good results in PD diagnosis but low sensitivity and PPV in distinguishing between these 2 diseases.^7,8 Limitations to the usefulness of transcranial sonography as a biomarker are its heavy dependence on the examiner’s experience and technical skills and the low sensitivity in differentiating between PD and PSP. Thus, reliability of transcranial sonography has not yet been established, especially for differentiation of PD from atypical parkinsonism. Some MRI studies have investigated the third ventricle size in PSP and PD patients using automated volumetry, confirming the enlargement of this structure in PSP.^9,10 A recent study¹⁰ in PSP patients has demonstrated that a combination of volume changes within 1 year in the third ventricle, midbrain, and frontal lobe correlated with progression in clinical scales, suggesting that this imaging parameter has clinical relevance as a biomarker of disease progression and further supporting the involvement of the third ventricle in PSP. Automated ventricular volumetry, however, is a time-consuming procedure that requires advanced technology and expertise, limiting the usefulness of this technique in clinical practice. Recently, we described a procedure for measuring the third ventricle width on an axial slice generated at the level of both the anterior and posterior commissures, normalized by the largest width of the frontal horns.⁶ This linear measurement was included in the MRPI calculation and led to a new biomarker, termed MRPI 2.0, which was highly accurate and reproducible in differentiating PD from PSP but not easy to perform and seems more suitable for research purposes than clinical routine.⁶ To maximize the feasibility of the 3^rdV quantitative measurement in clinical practice, in this study we developed a new simple manual MRI measurement of 3^rdV width, which can be performed on 3-D T1-weighted axial images generated by a standard subcallosal line, a procedure commonly used in radiologic routine.²³ This new imaging biomarker shows very high intra- and interrater reproducibility, does not require an automated procedure, and can be easily incorporated into routine MRI evaluation helping clinicians to better classify PD and PSP patients not only in research centers but also in clinical practice.

In accordance with previous reports,^6–10 our results showed that the enlargement of the third ventricle was a common finding in PSP but not in PD. In addition, we also demonstrated that disease progression did not influence the third ventricle width in PD patients because the 3^rdV/ID ratio did not vary over the 4-year follow-up despite the disease progression, as reflected by the increase in clinical scores. Our study showed a strong association between 3^rdV/ID value and PSP diagnosis in both the training and testing cohorts, suggesting that the higher the 3^rdV/ID, the higher the probability of having PSP. In line with this, the ROC curve showed that the 3^rdV/ID ratio yielded high diagnostic performance (AUC > 91% and accuracy > 87%) in differentiating PD from PSP patients in both cohorts. In fact, the results obtained in the training cohort were confirmed in a large independent international testing cohort, thus validating and generalizing the performance of our MR biomarker. In addition, similar 3^rdV/ID findings were also obtained when the optimal cutoff value of the training cohort was used in the testing cohort and vice versa. This latter approach strengthened the validity of our results, demonstrating the generalizability of cutoff values to new cohorts, enabling clinicians to use this biomarker in their patient sets.

It is worth highlighting that our MR biomarker performed well in the training cohort, which included late-stage patients, and in the testing cohort including de novo PD and PSP patients within the first years after disease onset, demonstrating its usefulness also in early-stage patients when clinical signs are mild, the levodopa responsiveness is less definite, and the diagnosis is still uncertain. This is particularly important because correctly classifying patients with parkinsonism is much more challenging at the beginning of the disease.^1–3

Our findings strongly suggest that in patients suspected of having PSP, the presence of an enlarged third ventricle (3^rdV/ID ≥ 5.88) supports the clinical diagnosis, whereas a small third ventricle representing a mismatch between clinical and imaging features suggests the need for further diagnostic evaluation or follow-up. In contrast, in de novo PD, the presence of a small third ventricle (3^rdV/ID < 5.88) helps to consolidate the clinical diagnosis, whereas an enlargement of this brain structure raises the suspicion of an alternative diagnosis, such as PSP or normal pressure hydrocephalus, 2 diseases with similar clinical and imaging phenotypes, which can be accurately differentiated using imaging biomarkers.²⁵ We thus suggest that patients suspected of having PD with an enlarged third ventricle should undergo a more extensive diagnostic workup and should not be included in clinical trials of putative neuroprotective and new potential disease-modifying therapies.

Our study provides new insights into the biomarkers’ selection criteria in parkinsonism,²⁶ suggesting the need for a balance between complexity and feasibility to distinguish biomarkers potentially suitable for research or clinical practice.

In the last decade, several biomarkers for differentiation of PSP and PD have been reported.⁵ MRPI and MRPI 2.0 have proven to be highly accurate in differentiating these 2 diseases in the early stages.^6,13 The main limitation to the wide use of these biomarkers, especially in clinical practice, is their complexity, which needs expertise in MR image reconstruction and in the manual measurement of the different small brain structures included in the calculation of these indexes.⁶ To overcome this limitation, an automated version of MRPI was recently developed by our group¹³ (an automated version of MRPI 2.0 is under development), but the software still requires some technical skills and is not suitable for routine MRI diagnostic procedures. Therefore, there is an urgent need for reliable, simple, and generalizable diagnostic biomarkers that can be performed on routine MR images by nonexpert physicians to be widely used in a clinical setting.²⁶ In the last few years, some simple MR planimetric measures have been proposed for differentiation of PSP from PD.^4,5 Of these, the most suitable for clinical practice were some linear measurements such as midbrain diameter and pons-to-midbrain diameter ratio.^27,28,29 These measures, however, lack generalizability because they have been tested in small patient groups, typically from a single center.^27,28 When used alone for differentiation between PSP and PD, these linear measures often showed low sensitivity not always meeting the 80% cut point required for an accurate biomarker,⁵ whereas they yielded high accuracy in differentiating PSP from non-PSP patients when included together with other measurements in decision tree algorithms.²⁹

In contrast, the 3^rdV/ID coupled simplicity with generalizability and good accuracy, thus enabling patients with PD or PSP to also be correctly classified in everyday clinical practice worldwide. Moreover, the use of our new MR biomarker could positively impact the enrollment of PD and PSP patients in clinical trials still based on clinical criteria. Indeed, the combined use of simple, generalizable biomarkers in addition to clinical criteria could allow more accurate selection of these patients across all centers. This new strategy could lead to more homogeneous patient groups and more reliable results in clinical trials, with a positive impact on the evaluation of the new disease-modifying therapies.

This study has several strengths. First, PD and PSP patients belonged to 2 large and independent cohorts with different stages of the disease (an Italian cohort of treated PD patients with mean disease duration of 3.6 ± 3.5 years and late-stage PSP patients with mean disease duration of 6.4 ± 2.9 years and an international cohort of de novo PD with mean disease duration of 0.5 ± 0.7 years, and early-stage PSP patients with mean disease duration of 2.2 ± 0.8 years), thus ensuring the generalizability of the results and also their feasibility in the early stage of the diseases. Second, all PD patients from both cohorts underwent a clinical and radiological 4-year follow-up, and only PD patients who maintained initial diagnosis at follow-up evaluation were enrolled to minimize the possibility of misclassification. Third, we developed a very simple linear measurement to assess third ventricle size (3^rdV/ID ratio) that is easy to perform, requires only a few minutes and little expertise, shows high reproducibility, and can be incorporated into routine MRI evaluation.

There are some limitations to this study. First, PSP and PD patients did not undergo a pathologic examination, and the clinical diagnosis might be in error in some patients. However, to reduce the possibility of misdiagnosis, we only enrolled PD patients who were followed-up for 4 years and PSP patients who were diagnosed according to international criteria with high sensitivity and specificity.³⁰ Second, PSP patients had higher cognitive impairment and were older than PD patients. Thus. we corrected the analysis for age, disease duration, and MMSE score. However, a minimal contribution of these variables in some of the between-group differences cannot be excluded. The correction for MMSE was only performed for the training cohort because PD and PSP patients in the testing cohort underwent different cognitive evaluations. Because the correction for MMSE did not modify the results in the training cohort, we believe that a lack of correction for cognitive data of PD and PSP in the testing cohort does not affect the validity of the results. Third, in our study in the testing cohort MRI images were obtained on different 1.5- and 3.0-T scanners with potential increased variability in 3^rdV/ID values. To minimize potential methodological bias, we corrected the analyses for the different scanner field strengths. Finally, it would be interesting to investigate the potential ability of our biomarker to differentiate PSP-P from PD in the early stage of the disease. However, the sample size of early PSP-P in the testing cohort was small, and further studies in a larger cohort of early PSP-P patients are warranted.

In conclusion, our study provides a new generalizable imaging measurement to differentiate PD from PSP patients using geographically different cohorts, also promising to be useful with data from new sites. Our results are relevant in the clinical practice because they provide physicians a new reliable MR biomarker that is easy to calculate on routine MR images for differentiation between PD and PSP in the early stages of the disease, when clinical diagnosis can still be uncertain. Finally, these data provide a strong impetus for using simple biomarkers in addition to clinical criteria in the selection of PD and PSP patients in clinical trials on disease-modifying treatments.