Over the past two decades multiple MRI markers, based on the selective midbrain and SCP atrophy of PSP, have been introduced in an effort to increase diagnostic accuracy of this rare disease. These markers initially included simple distance, area and volume measurements but have since evolved into more complex composite markers which incorporate ≥ 2 measurements, such as the P/M ratio and the MRPI. The reported diagnostic accuracy of most of these markers is based on limited studies of small cohorts of Parkinsonian disorders, often with poor representation of the major atypical Parkinsonian disorders (i.e., MSA, PSP, CBD). In addition, these cohorts exhibit great heterogeneity regarding clinical presentation, level of diagnostic certainty and disease duration of PSP patients among others. We aimed to systematically compare the diagnostic accuracy of available PSP MRI markers and examine the possible effects of disease duration, clinical presentation and level of diagnostic certainty in the performance of these imaging markers as surrogate markers of PSP.

An important initial finding of our study was the superiority of surface-based over distance-based and volume-based MRI markers in discriminating PSP from non-PSP patients. More specifically, midbrain surface exhibited higher accuracy (AUC = 0.937) over midbrain distance (AUC = 0.889) and volume (AUC = 0.819). Limited data exists regarding the direct comparison of linear, planimetric and volumetric MRI markers. An initial study supported the superiority of midbrain area over midbrain volume in differentiating PSP from CBD patients [12]. This superiority was further supported by a subsequent study reporting high (> 90%) diagnostic accuracy of midbrain surface compared to midbrain volume and distance, in a cohort of PSP and MSA patients [13]. This finding is in accordance with a recent meta-analysis of MRI markers in parkinsonian disorders, which supported the superiority of planimetric over volumetric MRI markers [25].

Interestingly, midbrain surface was superior to all composite MRI markers (P/M, P/M 2.0, MRPI, MRPI 2.0) in differentiating PSP from non-PSP patients in our cohort. Studies directly comparing the diagnostic accuracy of midbrain area to composite MRI markers are lacking. A single study focusing on the differentiation of PSP from PD/MSA patients reported superior diagnostic accuracy of midbrain surface (90%) compared to P/M ratio (81.4%) and MRPI (82.9%) [23]. A meta-analysis on PSP imaging markers also supported the superiority of midbrain surface over composite MRI markers [25].

The development of composite MRI markers (which include pontine morphometric measurements) aimed at differentiating PSP (characterized by predominant midbrain/SCP atrophy) from MSA (characterized by predominant pons/MCP atrophy). Multiple studies however have reported a concomitant moderate degree of pontine atrophy in PSP, as opposed to PD and control subjects, where pontine atrophy is minimal or absent [5, 7, 18,19,20]. Thus, the inclusion of pontine measurements in P/M and MRPI may in actuality result in a decrease in diagnostic accuracy, when differentiating PSP from PD or CBD patients. This point is of great clinical importance, since it supports the use of midbrain surface over the P/M ratio and MRPI as the optimal MRI marker in differentiating PSP from other parkinsonian disorders (with the possible exception of MSA).

Volume-based measurements performed poorer than surface- and distance-based measurements in discriminating PSP patients. None of the volume-based marker reached the threshold of 80% sensitivity and specificity. The sub-optimal diagnostic accuracy of volumetric MRI markers has been previously reported [12, 13, 15]. Volumetric measurements can only be performed automatically through dedicated software in clinical practice. Thus, limitations of volumetric MRI measurements rely heavily on the imaging processing package used and cannot be generalized. For the same reason, volumetric MRI data cannot be compared across different studies. Thus, the limitations of volumetric measurements reported herein apply specifically to the Freesurfer package which was utilized in this study and cannot be generalized for other packages analyzing structural neuroimaging data. Regarding linear measurements, a non-brainstem-based linear composite MRI marker (third ventricle width/internal skull diameter) has provided high diagnostic accuracy in differentiating of PSP from PD patients, indicating that morphometric MRI markers of other brain regions may assist in the diagnosis of Parkinsonian disorders [33].

The diagnostic accuracy of MRI markers depended on the level of diagnostic certainty in our cohort, with all MRI markers performing significantly better in cases of probable PSP vs. possible/suggestive PSP patients. Multiple markers provided high (> 80%) combined sensitivity/specificity in discriminating probable PSP patients, with diagnostic accuracy decreasing significantly when the same markers were applied in PSP patients fulfilling possible or suggestive criteria for PSP. Most studies in the field have analyzed mixed (i.e., probable and possible) PSP cohorts [5, 7, 12, 15, 17,18,19, 21, 23]. Data on differences between possible vs. probable PSP patients are scarce. A single study on the subject agreed with the superior performance of imaging markers in cohorts of probable vs. possible PSP patients [20]. This finding implies that MRI markers may have suboptimal accuracy in cases where there is clinical uncertainty on the diagnosis, which is the context where these markers are most relevant.

MRI markers performed better in PSP-RS patients compared to PSP variants in our cohort, as previously reported [34]. Several studies have supported the suboptimal diagnostic accuracy of MRI markers in PSP-P cohorts [7, 24, 35, 36]. To this end, novel MRI markers were introduced (i.e., MRPI 2.0 and M/P 2.0). These novel MRI markers produced suboptimal diagnostic accuracy on our PSP variant cohort, which however mainly consisted of PSP-F, PSP-CBS and PSP-SL patients. This indicates that although MRPI 2.0 is a potent imaging marker for PSP-P, it does not assist in the diagnosis of other PSP variants. Of interest, midbrain surface was the only MRI marker which provided high diagnostic accuracy in discriminating PSP variants and could be utilized in this context.

When analyzing MRI data among different PSP phenotypes, differences emerged, with PSP-F in our cohort exhibiting midbrain atrophy numerically comparable to PSP-RS. Previous studies have supported this finding, indicating that among PSP variants, PSP-F may present the greatest imaging similarities to PSP-RS [35]. However, due to the small number of patients per PSP phenotype, safe conclusions on this subject cannot be reached. Further studies are needed to replicate these findings.

An additional finding of this study was the effect of disease duration on the diagnostic accuracy of MRI markers. Importantly, there were no significant differences in the performance of these markers when patients were stratified based on disease duration. Even when applied in PSP patients with a disease duration of ≤ 24 months, all MRI markers retained their diagnostic accuracy. In addition, there were no significant differences regarding optimal cut-off values of MRI markers and which MRI markers were optimal for PSP diagnosis when cohorts of different disease duration were analyzed. Several studies support this finding. A study including MRI scans performed prior to disease onset in a small cohort of PSP patients indicated that midbrain atrophy precedes clinical manifestations [37]. Another study indicated that baseline MRI morphometric characteristics can can identify patients with unclassified Parkinsonism who develop PSP at follow-up [38]. These findings are clinically relevant, since they imply that morphometric changes present early in the disease course of PSP patients, and thus MRI markers can be applied in PSP patients in early stages of their disease course.

Lastly, we looked at possible correlations between midbrain surface (as the optimal MRI marker for PSP) and clinical characteristics of PSP patients, based on PSPRS total score and domain sub-scores. Midbrain surface correlated with total PSPRS, which is indicative of overall severity of symptoms. In addition, midbrain surface is highly correlated with the PSPRS ocular motor domain, gait domain and bulbar domain sub-scores. There were no correlations between midbrain atrophy and limb domain and cognitive domain subscores. Previous studies have highlighted the correlation between midbrain atrophy and ocular-motor dysfunction [39, 40]. These findings imply that midbrain atrophy is preferentially related to symptoms characteristic of PSP-RS (i.e., ocular motor dysfunction, gait difficulties) and does not correlate with symptoms more commonly present in PSP variants (limb and cognitive subscores). As discussed previously, midbrain atrophy performed significantly better in PSP-RS vs. PSP variants, which is in agreement with these findings.

This study has certain limitations. An initial limitation is the lack of neuropathological confirmation of clinical diagnoses, which is expected in studies of exceedingly rare disorders. This is the major limitation of all relevant studies on MRI markers in Parkinsonian disorders, with only few studies in the field including small cohorts of neuropathologically confirmed patients. These studies have provided conflicting results regarding the diagnostic accuracy of MRI markers in neuropathologically confirmed cases [35, 41,42,43]. This limitation is particularly important for patients with a CBD diagnosis. Although clinical diagnostic criteria for CBD have been established, these criteria lack neuropathological specificity [44]. Thus, it is likely that a subgroup of patients fulfilling criteria for CBD in our study may in fact harbor a different pathology, most commonly PSP or Alzheimer’s disease [45]. This is an inherent limitation of the established diagnostic criteria and should be taken into account in the interpretation of our study findings. This limitation extends to patients with a clinical diagnosis of PSP [46], particularly with a possible or suggestive level of diagnostic accuracy [47] and could result in an inflation of the diagnostic accuracy of MRI markers, as evidenced by studies reporting lower diagnostic accuracy [48]. In addition, data regarding the genetic status of our cohort is lacking. None of the patients included had a positive family history of an atypical parkinsonian disorder. However, it is possible that patients with a genetic form of atypical Parkinsonism may have been included in our cohort. PSP-P patients were underrepresented in this cohort. Our study only included patients who were hospitalized in our ward and did not include data from an outpatient clinic. Since PSP-P patients often manifest with asymmetrical tremor, responsive to dopaminergic treatment, they are commonly misdiagnosed as PD patients and are thus rarely hospitalized. In addition, the limitations of volumetric measurements reported herein cannot be generalized and refer to the Freesurfer package which was utilized throughout this study. Lastly, this study was retrospective and thus did not include longitudinal data which would provide more robust information on the temporal evolution of atrophy in PSP patients. Despite the retrospective inclusion of patients, data collection is being performed in a systematized, prospective manner in our ward, in an effort to have robust clinical data.

Notwithstanding these limitations, our study provides data based on one of the largest single-center cohorts to date, with an excellent representation of all major Parkinsonian disorders (CBD, MSA, PD). Importantly, all subjects underwent an identical, standardized MRI acquisition protocol and an automated MRI pre-processing protocol (for volumetry), which is pivotal for controlling for possible confounding factors present when utilizing different MRI acquisition protocols. To our knowledge, this study represents the first effort to systematically analyze the most commonly used MRI markers of different modality in a standardized manner. Along the same lines, it systematically examines the effects of clinical phenotype, level of diagnostic certainty and disease duration on the diagnostic accuracy of these MRI markers in the diagnosis of PSP, in an effort to better comprehend the limitations and applicability of these markers in clinical practice.

In conclusion, this single-center study supports the superiority of surface-based markers, and of midbrain surface in particular, among MRI markers in the diagnosis of PSP, even at early disease stages. However, the utility of MRI markers decreases significantly in patients with a possible/suggestive clinical diagnosis of PSP and in patients with PSP variants, which is a major limitation in the application of these markers in clinical practice. These limitations emphasize the primary importance of clinical diagnosis, which can be further supported by MRI markers in certain clinical settings. However, the findings of this study need further validation, taking into consideration the retrospective, single-center design of the study and the lack of neuropathological confirmation of diagnoses. A large-scale, longitudinal, multi-center study including neuropathological confirmation of the clinical diagnoses would provide more robust insights into the complex correlations of imaging and clinical phenomena in PSP patients throughout the natural course of this rare disorder.