A Neural Network-Based Approach to Hypothalamic Subdivision Classification
Poster No:
1747
Submission Type:
Abstract Submission
Authors:
Alexey Chervonnyy1, Eric Upschulte2, Sebastian Bludau2, Timo Dickscheid2, Katrin Amunts2
Institutions:
1Cécile and Oskar Vogt Institute of Brain Research, Düsseldorf, North Rhine-Westphalia, 2Forschungszentrum Juelich GmbH, Juelich, North Rhine-Westphalia
First Author:
Alexey Chervonnyy
Cécile and Oskar Vogt Institute of Brain Research
Düsseldorf, North Rhine-Westphalia
Cécile and Oskar Vogt Institute of Brain Research
Düsseldorf, North Rhine-Westphalia
Co-Author(s):
Introduction:
Image analysis and statistical methods have enabled the identification of borders between cortical areas by extracting line profiles to capture laminar cell density changes, resulting in reproducible anatomical maps of cortical areas and a better understanding of cortical organisation (1). However, subcortical structures present unique challenges due to their distinct organisation, characterised by nuclei, cell clusters and intersecting fibre tracts. This has hindered the development of methods to reliably confirm the distinctiveness of subcortical subdivisions, an important prerequisite for a more comprehensive understanding of the relationship between the structure of nuclei and their connectivity and function. To address this challenge, we combined texture analysis with a contour proposal network to create a neural network capable of reliably classifying hypothalamic subdivisions.
Methods:
Using every 15th cell body stained brain section (1 μm resolution) from 10 postmortem brains (5 female), including the BigBrain dataset (2), we delineated the hypothalamus and its nuclei. Texture analysis was performed on 6709 regions of interest derived from these delineations, employing the Gray Level Co-occurrence Matrix method (3) to quantify spatial relationships and intensity patterns in grayscale images. We extracted 21 modified Haralick texture features, such as correlation, homogeneity, and dissimilarity, to characterise the tissue's cellular architecture (4). Differences between hypothalamic subdivisions were assessed using the independent-samples Kruskal-Wallis test.
In parallel, neurons were segmented using a Contour Proposal Network based on Fourier Descriptors (5), enabling precise measurements of neuron number, size, and morphology. To enhance classification accuracy, we trained a one-layer neural network model using 186 predictors derived from texture and contour features. The model, with 10 ReLU-activated hidden neurons and a softmax classification layer, was trained on 5690 observations with 5-fold cross-validation.
In parallel, neurons were segmented using a Contour Proposal Network based on Fourier Descriptors (5), enabling precise measurements of neuron number, size, and morphology. To enhance classification accuracy, we trained a one-layer neural network model using 186 predictors derived from texture and contour features. The model, with 10 ReLU-activated hidden neurons and a softmax classification layer, was trained on 5690 observations with 5-fold cross-validation.
Results:
23 subdivisions of the hypothalamus were identified (Fig.1). Some, such as the paraventricular nucleus, showed a distinct cytoarchitecture with magnocellular neurons, while others, like the uncinate and intermediate nuclei, displayed similar features, consisting of medium-sized neurons.
Principal Component Analysis (PCA; SPSS v.29) identified four main components explaining 87.27% of the total variance. Significant differences in at least one main component were observed between all adjacent nuclei, supporting their delineation. For visualisation, we generated a heatmap (Fig.2) indicating levels of cytoarchitectural difference: a score of 0 showed no significant differences, while a score of 4 indicated pronounced disparities across all components.
In addition, some more distant nuclei, such as the uncinate and suprachiasmatic nuclei, showed no significant differences in the PCA components. These cytoarchitectural similarities may suggest functional connectivity between distant nuclei and warrant further investigation of their interactions.
The contour proposal network enabled pixel-level labeling of cells in microscopic images, facilitating the identification of individual neurons. Using the extracted data, such as the number of neurons and their size, we calculated the cell packing density and observed the highest density in the supraoptic nucleus and the lowest in the lateral tuberal nucleus, which was three times less dense.
By integrating numerical data from texture analysis with neuronal metrics obtained from the contour proposal network, our model classified hypothalamic subdivisions with 86.2% accuracy.
Principal Component Analysis (PCA; SPSS v.29) identified four main components explaining 87.27% of the total variance. Significant differences in at least one main component were observed between all adjacent nuclei, supporting their delineation. For visualisation, we generated a heatmap (Fig.2) indicating levels of cytoarchitectural difference: a score of 0 showed no significant differences, while a score of 4 indicated pronounced disparities across all components.
In addition, some more distant nuclei, such as the uncinate and suprachiasmatic nuclei, showed no significant differences in the PCA components. These cytoarchitectural similarities may suggest functional connectivity between distant nuclei and warrant further investigation of their interactions.
The contour proposal network enabled pixel-level labeling of cells in microscopic images, facilitating the identification of individual neurons. Using the extracted data, such as the number of neurons and their size, we calculated the cell packing density and observed the highest density in the supraoptic nucleus and the lowest in the lateral tuberal nucleus, which was three times less dense.
By integrating numerical data from texture analysis with neuronal metrics obtained from the contour proposal network, our model classified hypothalamic subdivisions with 86.2% accuracy.
Conclusions:
This integrated approach represents a significant advance in studying subcortical structures, improving the reproducibility and precision of histological analyses. These findings provide deeper insights into hypothalamic architecture and lay a foundation for exploring its functional and structural connectivity.
Neuroanatomy, Physiology, Metabolism and Neurotransmission:
Cortical Anatomy and Brain Mapping 2
Subcortical Structures 1
Keywords:
Segmentation
Statistical Methods
Structures
Sub-Cortical
Other - Hypothalamus
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
Provide references using APA citation style.
2. Amunts, K. (2013). BigBrain: An Ultrahigh-Resolution 3D Human Brain Model. Science, 340(6139), 1472- 1475.
3. Haralick, R. M. (1973). Textural Features for Image Classification. Man, and Cybernetics, vol. SMC-3, no. 6, pp. 610-621.
4. Löfstedt, T. (2019). Gray-level invariant Haralick texture features. PLOS ONE 14(2): e0212110.
5. Upschulte, E. (2022). Contour proposal networks for biomedical instance segmentation. Medical image analysis, 77, 102371.