Differentiating between common PSP phenotypes using structural MRI: a machine learning study

Added value of FDG-PET for detection of progressive supranuclear palsy

BACKGROUND

Diagnostic criteria for progressive supranuclear palsy (PSP) include midbrain atrophy in MRI and hypometabolism in [18F]fluorodeoxyglucose (FDG)-positron emission tomography (PET) as supportive features. Due to limited data regarding their relative and sequential value, there is no recommendation for an algorithm to combine both modalities to increase diagnostic accuracy. This study evaluated the added value of sequential imaging using state-of-the-art methods to analyse the images regarding PSP features.

METHODS

The retrospective study included 41 PSP patients, 21 with Richardson's syndrome (PSP-RS), 20 with variant PSP phenotypes (vPSP) and 46 sex- and age-matched healthy controls. A pretrained support vector machine (SVM) for the classification of atrophy profiles from automatic MRI volumetry was used to analyse T1w-MRI (output: MRI-SVM-PSP score). Covariance pattern analysis was applied to compute the expression of a predefined PSP-related pattern in FDG-PET (output: PET-PSPRP expression score).

RESULTS

The area under the receiver operating characteristic curve for the detection of PSP did not differ between MRI-SVM-PSP and PET-PSPRP expression score (p≥0.63): about 0.90, 0.95 and 0.85 for detection of all PSP, PSP-RS and vPSP. The MRI-SVM-PSP score achieved about 13% higher specificity and about 15% lower sensitivity than the PET-PSPRP expression score. Decision tree models selected the MRI-SVM-PSP score for the first branching and the PET-PSPRP expression score for a second split of the subgroup with normal MRI-SVM-PSP score, both in the whole sample and when restricted to PSP-RS or vPSP.

CONCLUSIONS

FDG-PET provides added value for PSP-suspected patients with normal/inconclusive T1w-MRI, regardless of PSP phenotype and the methods to analyse the images for PSP-typical features.

Multimodal imaging and electrophysiological study in the differential diagnosis of rest tremor

Introduction

Distinguishing tremor-dominant Parkinson's disease (tPD) from essential tremor with rest tremor (rET) can be challenging and often requires dopamine imaging. This study aimed to differentiate between these two diseases through a machine learning (ML) approach based on rest tremor (RT) electrophysiological features and structural MRI data.

Methods

We enrolled 72 patients including 40 tPD patients and 32 rET patients, and 45 control subjects (HC). RT electrophysiological features (frequency, amplitude, and phase) were calculated using surface electromyography (sEMG). Several MRI morphometric variables (cortical thickness, surface area, cortical/subcortical volumes, roughness, and mean curvature) were extracted using Freesurfer. ML models based on a tree-based classification algorithm termed XGBoost using MRI and/or electrophysiological data were tested in distinguishing tPD from rET patients.

Results

Both structural MRI and sEMG data showed acceptable performance in distinguishing the two patient groups. Models based on electrophysiological data performed slightly better than those based on MRI data only (mean AUC: 0.92 and 0.87, respectively; p = 0.0071). The top-performing model used a combination of sEMG features (amplitude and phase) and MRI data (cortical volumes, surface area, and mean curvature), reaching AUC: 0.97 ± 0.03 and outperforming models using separately either MRI (p = 0.0001) or EMG data (p = 0.0231). In the best model, the most important feature was the RT phase.

Conclusion

Machine learning models combining electrophysiological and MRI data showed great potential in distinguishing between tPD and rET patients and may serve as biomarkers to support clinicians in the differential diagnosis of rest tremor syndromes in the absence of expensive and invasive diagnostic procedures such as dopamine imaging.

Combined cortical thickness and blink reflex recovery cycle to differentiate essential tremor with and without resting tremor

Objective

To investigate the performance of structural MRI cortical and subcortical morphometric data combined with blink-reflex recovery cycle (BRrc) values using machine learning (ML) models in distinguishing between essential tremor (ET) with resting tremor (rET) and classic ET.

Methods

We enrolled 47 ET, 43 rET patients and 45 healthy controls (HC). All participants underwent brain 3 T-MRI and BRrc examination at different interstimulus intervals (ISIs, 100-300 msec). MRI data (cortical thickness, volumes, surface area, roughness, mean curvature and subcortical volumes) were extracted using Freesurfer on T1-weighted images. We employed two decision tree-based ML classification algorithms (eXtreme Gradient Boosting [XGBoost] and Random Forest) combining MRI data and BRrc values to differentiate between rET and ET patients.

Results

ML models based exclusively on MRI features reached acceptable performance (AUC: 0.85-0.86) in differentiating rET from ET patients and from HC. Similar performances were obtained by ML models based on BRrc data (AUC: 0.81-0.82 in rET vs. ET and AUC: 0.88-0.89 in rET vs. HC). ML models combining imaging data (cortical thickness, surface, roughness, and mean curvature) together with BRrc values showed the highest classification performance in distinguishing between rET and ET patients, reaching AUC of 0.94 ± 0.05. The improvement in classification performances when BRrc data were added to imaging features was confirmed by both ML algorithms.

Conclusion

This study highlights the usefulness of adding a simple electrophysiological assessment such as BRrc to MRI cortical morphometric features for accurately distinguishing rET from ET patients, paving the way for a better classification of these ET syndromes.

Periodic Leg Movements during Sleep Associated with REM Sleep Behavior Disorder: A Machine Learning Study

Most patients with idiopathic REM sleep behavior disorder (iRBD) present peculiar repetitive leg jerks during sleep in their clinical spectrum, called periodic leg movements (PLMS). The clinical differentiation of iRBD patients with and without PLMS is challenging, without polysomnographic confirmation. The aim of this study is to develop a new Machine Learning (ML) approach to distinguish between iRBD phenotypes. Heart rate variability (HRV) data were acquired from forty-two consecutive iRBD patients (23 with PLMS and 19 without PLMS). All participants underwent video-polysomnography to confirm the clinical diagnosis. ML models based on Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost) were trained on HRV data, and classification performances were assessed using Leave-One-Out cross-validation. No significant clinical differences emerged between the two groups. The RF model showed the best performance in differentiating between iRBD phenotypes with excellent accuracy (86%), sensitivity (96%), and specificity (74%); SVM and XGBoost had good accuracy (81% and 78%, respectively), sensitivity (83% for both), and specificity (79% and 72%, respectively). In contrast, LR had low performances (accuracy 71%). Our results demonstrate that ML algorithms accurately differentiate iRBD patients from those without PLMS, encouraging the use of Artificial Intelligence to support the diagnosis of clinically indistinguishable iRBD phenotypes.

Planimetric and Volumetric Brainstem MRI Markers in Progressive Supranuclear Palsy, Multiple System Atrophy, and Corticobasal Syndrome. A Systematic Review and Meta-Analysis

BACKGROUND

Various MRI markers-including midbrain and pons areas (Marea, Parea) and volumes (Mvol, Pvol), ratios (M/Parea, M/Pvol), and composite markers (magnetic resonance imaging Parkinsonism Indices 1,2; MRPI 1,2)-have been proposed as imaging markers of Richardson's syndrome (RS) and multiple system atrophy-Parkinsonism (MSA-P). A systematic review/meta-analysis of relevant studies aiming to compare the diagnostic accuracy of these imaging markers is lacking.

METHODS

Pubmed and Scopus were searched for studies with >10 patients (RS, MSA-P or CBS) and >10 controls with data on Marea, Parea, Mvol, Pvol, M/Parea, M/Pvol, MRPI 1, and MRPI 2. Cohen's d, as a measure of effect size, was calculated for all markers in RS, MSA-P, and CBS.

RESULTS

Twenty-five studies on RS, five studies on MSA-P, and four studies on CBS were included. Midbrain area provided the greatest effect size for differentiating RS from controls (Cohen's d = -3.10; p < 0.001), followed by M/Parea and MRPI 1. MSA-P had decreased midbrain and pontine areas. Included studies exhibited high heterogeneity, whereas publication bias was low.

CONCLUSIONS

Midbrain area is the optimal MRI marker for RS, and pons area is optimal for MSA-P. M/Parea and MRPIs produce smaller effect sizes for differentiating RS from controls.