Related Experiment Video
Updated: Mar 13, 2026
Automated Segmentation of Cortical Grey Matter from T1-Weighted MRI Images
Published on: January 7, 2019
Hypothalamic atrophy in progressive supranuclear palsy, assessed by convolutional neural network-based automatic
Jan Kassubek1, Günter U Höglinger2,3,4, Adam Zůza5
1Department of Neurology, University of Ulm, Oberer Eselsberg 45, 89081, Ulm, Germany. jan.kassubek@uni-ulm.de.
Background:
The hypothalamus as one of the core structures in metabolic control is increasingly recognized to be morphologically altered in various neurodegenerative diseases.
Objective:
The purpose of this study was to quantitatively investigate the hypothalamic volumes in patients with progressive supranuclear palsy (PSP) and to compare them with controls and Parkinson disease (PD) patients.
Methods:
An automatic hypothalamic volume quantification method based on the use of convolutional neural networks (CNN) of U-Net architecture was applied to the automatic segmentation of the hypothalamus and intracranial volumes (ICV). This CNN-based volumetric analysis was performed in high resolution T1 weighted MRI in two PSP cohorts: cohort A with 78 PSP patients and 63 controls was recorded at 3.0 T at multiple sites; the single site cohort B consisted of 66 PSP patients, 66 PD patients, and 44 controls, recorded at 1.5 T.
Results:
In cohort A, significant hypothalamic volume reduction was observed in PSP (774 ± 83 mm3) when compared to controls (817 ± 74 mm3). In cohort B, this result of significant hypothalamic volume reduction was confirmed in PSP (745 ± 102 mm3) when compared to controls (831 ± 81 mm3); no significant hypothalamic volume reduction was observed in PD (797 ± 98 mm3), in support of previous studies.
Conclusion:
The CNN-based hypothalamus volume quantification study demonstrated significantly reduced hypothalamus volumes in PSP patients compared to controls and PD, respectively; future studies will address the metabolic profiles of PSP as potential functional correlates.
