Defining the human hypothalamus in vivo by ultra-high field 7 Tesla MRI

S Schindler; L Schmidt; M Strauß; A Anwander; PL Bazin; R Trampel; H Möller; U Hegerl; R Turner; S Geyer; P Schönknecht

doi:10.1055/s-0033-1337206

Klinische Neurophysiologie, Table of Contents

Defining the human hypothalamus in vivo by ultra-high field 7 Tesla MRI

S Schindler ¹, L Schmidt ¹, M Strauß ¹, A Anwander ², PL Bazin ², R Trampel ², H Möller ², U Hegerl ¹, R Turner ², S Geyer ², P Schönknecht ¹

¹Uniklinik, Psychiatrie, Leipzig, Deutschland
²Max Planck Institut für Kognitions und Neurowissenschaften, Leipzig, Deutschland

Abstract

Introduction:

Post mortem studies have shown a volume deficit of the hypothalamus in depressive patients. With ultra-high field (7 Tesla) MRI this effect can now be investigated in vivo in detail. But to benefit from the sub-millimeter resolution a segmentation procedure was required that overcomes limitations of existing procedures, in particular schematic approximations.

Methods:

Using 7 Tesla T1 images the traditional anatomical landmarks of the hypothalamus were refined. A detailed segmentation algorithm (unilateral hypothalamus) was developed for colour coded, histogram-matched images and evaluated in a sample of 10 subjects. Intra- and interrater reliabilities were estimated in terms of intra- and interclass-correlation coefficients (ICC).

Results:

The computer-assisted segmentation algorithm ensured test-retest reliabilities of ICC ≥0.97 and interrater-reliabilities of ICC ≥0.94. There were no significant volume differences between the tracers and between the hemispheres (paired t-tests). The estimated volume of the hypothalamus (tracer 1, first run) was 1130.64 mm³± 103.48 mm³.

Conclusion:

We present a computer-assisted algorithm for the manual segmentation of the human hypothalamus on ultra-high field 7 Tesla MR images. With very high intra- and interrater reliabilities it outperforms former procedures established with 1.5 T or 3 T MRI. The estimated volumes lie between previous histological and neuroimaging results. The algorithm provides an excellent basis for the investigation of our larger neuropsychiatric sample. It can be used by fellow researchers and it can serve as a gold standard for future automated procedures.