H-SynEx: Using synthetic images and ultra-high resolution ex vivo MRI for hypothalamus subregion segmentation

TL;DR

This work tackles the challenge of automated hypothalamic subregion segmentation across diverse MRI sequences and resolutions. It introduces H-SynEx, a two-stage segmentation framework trained on synthetic images derived from ultra-high-resolution ex vivo label maps, enabling cross-sequence generalization without retraining. The approach combines a whole-structure segmentation model and a subregion model, optimized with tailored loss functions, and validated across six in vivo datasets and multiple MRI sequences, including FLAIR with 5 mm spacing. Results show robust segmentation performance, the ability to discriminate patient groups from controls in AD and bvFTD, and public availability of the method, highlighting its potential for broader neurodegenerative research and clinical application.

Abstract

The hypothalamus is a small structure located in the center of the brain and is involved in significant functions such as sleeping, temperature, and appetite control. Various neurological disorders are also associated with hypothalamic abnormalities. Automated image analysis of this structure from brain MRI is thus highly desirable to study the hypothalamus in vivo. However, most automated segmentation tools currently available focus exclusively on T1w images. In this study, we introduce H-SynEx, a machine learning method for automated segmentation of hypothalamic subregions that generalizes across different MRI sequences and resolutions without retraining. H-synEx was trained with synthetic images built from label maps derived from ultra-high resolution ex vivo MRI scans, which enables finer-grained manual segmentation when compared with 1mm isometric in vivo images. We validated our method using Dice Coefficient (DSC) and Average Hausdorff distance (AVD) across in vivo images from six different datasets with six different MRI sequences (T1, T2, proton density, quantitative T1, fractional anisotrophy, and FLAIR). Statistical analysis compared hypothalamic subregion volumes in controls, Alzheimer's disease (AD), and behavioral variant frontotemporal dementia (bvFTD) subjects using the Area Under the Receiving Operating Characteristic curve (AUROC) and Wilcoxon rank sum test. Our results show that H-SynEx successfully leverages information from ultra-high resolution scans to segment in vivo from different MRI sequences. Our automated segmentation was able to discriminate controls versus Alzheimer's Disease patients on FLAIR images with 5mm spacing. H-SynEx is openly available at https://github.com/liviamarodrigues/hsynex.

Figures (10)

• Figure 1: ex vivo MR images: Examples of three images used during the method development
• Figure 2: Example of different modalities (FSM dataset)
• Figure 3: Image preprocessing and label maps creation. (a) Original ex vivo image. (b) Preprocessed image (c) Automated brain segmentation ($k$=4) and manual hypothalamus segmentation merged (d) The final label map is cropped around the hypothalamus (yellow box) to generate the synthetic images.
• Figure 4: Examples of coronal slices from 3D synthetic images used as input: The images shown here came from the label maps cropped around the hypothalamus. The use of aggressive data augmentation along random contrast values on the generative model results in large variability in the appearance of the input images.
• Figure 5: Training Flowchart: (a) Generation of synthetic images: The synthetic images S are generated using the label maps from the ex vivo images. (b) Models training: there are two training blocks, one focused on the entire hypothalamus and another specialized in subregion segmentation. The training of the two blocks is done subsequently. We first trained the whole structure segmentation model($M_{hyp}$), and later, the model for the subregions segmentation($M_{sub}$). However, the output of $M_{hyp}$ is used to assist the input creation of $M_{sub}$ during training.