14:30   Brain & Eye Chair: Robert-Jan Doll Integrating fixel based analysis and lengthwise profiling for assessing changes in white matter tracts Lloyd Plumart, Hinke Halbertsma, Mayra Bittencourt, Remco Renken, Frans Cornelissen Abstract: Diffusion weighted imaging (DWI) is an imaging modality within magnetic resonance imaging (MRI) that can be used to describe microstructural properties of highly diffusive structures, such as the white matter of the brain[1]. A promising approach to analyze DWI data and describe white matter integrity is fixel based analysis (FBA). FBA characterizes different fiber populations within a voxel, called ‘fixels’, by computing their microstructural metrics, such as fiber density (FD), fiber cross-section (FC) and a combined metric (FDC). Reductions in the first two metrics are associated with axonal loss and atrophy, respectively[2]. Recent studies applying FBA to DWI data of people with the ophthalmic neurodegenerative disease glaucoma have suggested a propagation of degeneration along the primary visual pathway (i.e. the anterior optic pathway (AOP) and optic radiation (OR))[3]. In this study, we developed a lengthwise profiling algorithm and combined it with FBA to more precisely visualize and quantify how neurodegeneration propagates. We then applied this profiling technique to the primary visual pathway in glaucoma. However, this technique can be applied to any tract of interest to more accurately quantify neurodegeneration at the tract level, compared to traditional FBA. We acquired DWI data from 17 patients with glaucoma (mean age 70.8 ± 8.7 years) and 26 age-matched controls (mean age 68.5 ± 7.7 years) using 3T MRI. The recommended FBA pipeline of MRtrix3 was followed[4], after which the primary visual pathway was tracked using tractography. Lengthwise profiling was performed by determining the core of each tract (AOP and OR) and projecting the surrounding fixel values onto that core tract, thereby creating FD, FC, and FDC profiles along the tract. Compared to healthy controls, the glaucoma group showed significant reductions for all FBA metrics along the entire visual pathway. Furthermore, the FD, FC, and FDC profiles captured the progressive downstream degeneration of the visual pathway in glaucoma. By constructing detailed lengthwise spatial profiles of FD, FC, and FDC along tracts of interest, this integrated analysis yields a promising approach to monitor disease progression and evaluate the effects of novel treatments for neurodegenerative diseases. Enhancing 3D US Fetal Brain Segmentation Through MRI-Based Label Transfer Gaby van Iersel, Jalmar Teeuw, Inge van Ooijen, Mireille Bekker, Manon Benders, Ruud van Sloun, Hilleke Hulshoff Pol, Sonja de Zwarte Abstract: Introduction: Fetal brain imaging is crucial for prenatal diagnosis, with MRI providing high-resolution images and ultrasound (US) being widely accessible. However, labeling US images accurately is challenging due to low soft tissue contrast and artifacts. This study explores transferring MRI labels to US images to improve fetal US brain segmentation. The preliminary results in this abstract focus on total brain volume (TBV) segmentation. Methods: 3D Data from the YOUth Baby & Child US & MRI dataset was used, focusing on 15 subjects with MRI and US scans within 24 hours. After excluding poor-quality scans, 45 US scans (N = 12 subjects) were analyzed. MRI scans were pre-labeled with 19 anatomical structures using the BOUNTI tool. US and MRI data were manually registered using ITKsnap software. A V-Net-based segmentation network was used for TBV segmentation (3 downsampling and upsampling stages). The dataset was augmented and split into training, validation, and test sets with ratio of 15:17:15 US scans (N = 4:4:4). The model performance was evaluated using test time augmentation, to calculate the mean Dice Score (DS), Hausdorff distance (HD), and Center of Mass Distance (COM). Results: The evaluation metrics of TBV segmentation showed promising results, with a mean DS of 0.894 (std ± 0.023) and COM deviations with a mean of 3.708 voxels (std ± 1.489). Although the HD was high (mean 24.237 voxels, std ± 12.244), the COM is of greater importance as this will be used to crop the US images. The cropped region will serve as the input for a subsequent network focused on substructure segmentation. Discussion: V-Net-based segmentation network work well for TBV segmentation. Future work will focus on US alignment and training a network for brain substructure segmentation. We also plan to apply our model to the complete YOUth dataset of around 50,000 scans (N=2777 subjects) at 20 and 30 weeks of gestation to study brain development. The goal is to develop a robust tool for fetal brain segmentation that uses the benefits of MRI quality while minimizing reliance on extensive MRI datasets. This approach could advance fetal brain development research and improve prenatal diagnosis. Impact of Diurnal Effects on EEG Power in Resting-State Pharmaco-EEG Studies: A Meta-Analysis Naser Hakimi, Annika A. de Goede, Geert Jan Groeneveld, Robert-Jan Doll Abstract: Background: Quantitative pharmaco-EEG (pEEG) is an important tool for evaluating drug effects on the central nervous system, with the absolute power of EEG frequency bands serving as a reliable biomarker for these effects[1]. Understanding diurnal effects on EEG is important to distinguish between drug-induced changes and natural fluctuations in brain activity. We aim to investigate diurnal effects on EEG power in resting-state pEEG. Methods: This meta-analysis used data from 646 unique healthy participants enrolled in a total of 24 clinical trials at our research center. All trials included two pre-dose (baseline) recordings and several post-drug administration recordings, spanning approximately from 8 a.m. to 10 p.m. The resting-state protocol consisted of 5 alternating 64-seconds periods of eyes open and closed, with Fz-Cz and Pz-Oz as electrodes of interest. The analysis was divided into two parts: (1) a comparison of EEG absolute power between the two baseline recordings, and (2) an analysis of diurnal effects on EEG power using pre-dose and post-dose recordings from 258 placebo participants. Linear mixed-effects models were used, with time as a fixed factor, while accounting for subject variance as a random effect, for each electrode and frequency band. Results: Part 1 showed significantly increased power during the second baseline recording for all bands (p<.001). In part 2, EEG power demonstrated increasing powers, peaking around 3 p.m. in the delta and theta bands and stabilizing by 8 p.m. in the other bands. Discussion: We confirm that diurnal effects, likely influenced by factors such as fatigue, significantly affect EEG power, even within short intervals (i.e., <1 hour) observed in baseline recordings, with stronger increases in specific power bands in the morning compared to the afternoon. However, we cannot exclude the potential contributions of other factors such as signal quality. To avoid misattributing time-of-day EEG power variations to drug effects, future studies should account for them when designing protocols and analyzing data. [1] M. Jobert, F.J. Wilson, G.S. Ruigt, M. Brunovsky, L.S. Prichep, W.H. Drinkenburg, et al., Guidelines for the recording and evaluation of pharmaco-EEG data in man: the International Pharmaco-EEG Society (IPEG), Neuropsychobiology, 66 (2012), pp. 201-220 pEEG-based Bbrain age Iindex eEstimation Mmethods using clinically validated endpoints Mihir Kapadia, Naser Hakimi, Annika de Goede, Robert-Jan Doll Abstract: Background: The use of electroencephalography (EEG) to assess brain age and cognitive function has been a topic of interest in recent years. Brain age index refers to the deviation of estimated age from the chronological age obtained from neurological data, which is a potential biomarker of cognitive function [1]. Spectrogram-based Convolutional Neural Network (CNN) models have been used to predict the brain age index, highlighting the potential of EEG-based metrics in assessing brain health and function [2]. Furthermore, EEG has been used to compare EEG feature covariates to predict life expectancy and survival rate for terminal pathologies [3]. However, there is lack of a standardised measuring protocol and analysis method across different studies. We aim to investigate the use of pharmaco-EEG (pEEG) collected using a standardized protocol [4] to derive biomarkers for a reliable estimation of brain age index. Such a biomarker can act as a metric for measurement of pharmacodynamic effects of drugs in healthy as well as patient populations. Method: This meta-analysis used data from 646 participants enrolled in a total of 24 clinical trials. All trials included pre-dose (baseline) recordings and several post-drug administration recordings. The analysis consisted of using machine learning models for performing a regression analysis to estimate brain age. The regression analysis was performed on the EEG power band features within different frequency bands (delta, theta, alpha, beta, and gamma) at electrodes Fz-Cz and Pz-Oz. Preliminary results: Using ElasticNet Regression estimator, the preliminary model predicts age with a mean absolute error of 10.54 years, and a root mean squared error of 15.01 years across healthy participants. Discussion: Our preliminary findings align with those reported in the literature, utilizing reliable EEG biomarkers and standard protocols commonly employed in pEEG studies. This suggests that after further improvements of the model, the results of our study could be directly applied in pEEG research to assess the pharmacodynamic effects of drugs in neurodegenerative diseases. [1] Sun, H. et al. Brain age from the electroencephalogram of sleep. Neurobiol. Aging 74, 112–120 (2019). [2] Yook, S. et al. Novel neuroelectrophysiological age index associated with imaging features of brain aging and sleep disorders. Neuroimage 264, 119753 (2022). [3] Paixao, L. et al. Excess brain age in the sleep electroencephalogram predicts reduced life expectancy. Neurobiol. Aging 88, 150–155 (2020). [4] Jobert, M. et al. Guidelines for the recording and evaluation of pharmaco-eeg data in man: The international pharmaco-EEG society (IPEG): The IPEG pharmaco-EEG guideline committee. Neuropsychobiology 66, 201–220 (2012). Superpixel Approach for Intraoperative Brain Tumour Delineation in Hyperspectral Images Max Verbers, Francesca Manni, Svitlana Zinger Abstract: Brain cancer is the 12th most common cause of cancer mortality, with an estimate 308,000 cases and 251,000 deaths reported globally in 2020 [1]. In case of high grade glioma, a surgical resection is often performed. In this case, detecting the tumour boundaries for resection is challenging due to its invasiveness. Improving the surgical resection during brain tumour surgery can lead to decrease mortality rate, recurrence, and potential complications. This can be achieved by properly visualizing the lesion during surgery and safely resecting the tumour margin. The European project STRATUM aims to develop a 3D based-assistance tool to guide neurosurgeons in brain tumour resection surgery, utilizing Hyperspectral Imaging (HSI) for intraoperative tumour delineation. In order to apply HSI for intraoperative brain tumour detection and delineation, a dedicated image analysis framework should entail HSI data preprocessing, hyperspectral dimensionality reduction and classification. Reducing the high dimensionality of HSI data is crucial for intraoperative real-time processing and improving classification performance. We present an innovative HSI data processing framework for reducing dimensionality and extracting key features for an AI-based algorithm. This approach enhances robustness against spatial noise and eliminates redundant information [2], hence potentially improving the accuracy and efficiency of the surgical procedures. In our experiments, 24 images of 14 patients with grade IV brain tumours are analysed [3]. To segment the tumor and healthy areas, we applied patching and superpixel techniques. Preliminary results with a linear SVM show that the average superpixel and patching techniques reach F1 scores of 70.91% and 68.85% respectively, while using all pixels gives a score of 76.59%. The feature space was reduced 100 and 40 times for superpixel and patching techniques, respectively, with only a 5% performance loss. These findings suggest that spatial dimensionality reduction techniques can lead to high classification performances while reducing computational load, which is beneficial in a real-time intraoperative scenario. In conclusion, the results demonstrate that the hyperspectral feature space can be significantly reduced with minimal performance loss. Future work will focus on refining these techniques and combine them with spectral reduction approaches to further improve the performances of the automatic tumour detection algorithm. Population receptive field size across cortical depth along the visual hierarchy Mayra Bittencourt, Marcus Daghlian, Remco Renken, Serge Dumoulin, Frans W. Cornelissen Abstract: In the visual cortex, population receptive field (pRF) size increases both with eccentricity and when moving up along the visual hierarchy. Previous functional magnetic resonance imaging (fMRI) and neurophysiology studies found that in the primary visual cortex (V1), pRF size varies across cortical depth according to a U-shaped function, with the smallest pRF sizes in central layers. This U-shaped pattern is thought to reflect the hierarchical information flow across cortical depth, where the information arrives in central layers and is further processed in superficial and deeper layers. However, it is still unknown how pRF properties are organized across cortical depth in later visual areas. Here, we use population receptive field modeling at ultra-high field (7T) functional MRI to investigate pRF size variation across cortical depth and along the visual hierarchy (i.e. V1-hV4, LO-1 and LO-2) at sub-millimeter resolution (0.8mm isotropic). We acquired neuroimaging data from 13 healthy participants (mean age ± SD: 34 ± 9 years old, seven females), each of whom performed four to six sessions each of a visual retinotopy task. Functional data preprocessing included thermal denoising, susceptibility distortion correction, motion correction and high-pass filtering. Both anatomical and functional data were upsampled to 0.4mm isotropic resolution. Anatomical images were co-registered to functional images, segmented into gray matter, and divided into eight equivolumetric cortical surface layers. Our results show that in V1, pRF size follows the expected U-shaped function with cortical depth. In V2 and beyond, our preliminary results did not reveal an U-shaped function, potentially suggesting area specific associations between information flow and cortical areas. This study brings new evidence on the laminar organization of pRF properties along the visual hierarchy, which is relevant for the fundamental understanding of human (visual) brain function.

end %-->