Corrigendum to "Cross-sectional study of retrospective self-reported childhood emotional neglect and inhibitory neurometabolite levels in the pregenual anterior cingulate cortex in adult humans" [Neurobiol. Stress, 25 (July 2023), 100556].

[This corrects the article DOI: 10.1016/j.ynstr.2023.100556.].

VesselBoost: A Python Toolbox for Small Blood Vessel Segmentation in Human Magnetic Resonance Angiography Data.

Magnetic resonance angiography (MRA) performed at ultra-high magnetic field provides a unique opportunity to study the arteries of the living human brain at the mesoscopic level. From this, we can gain new insights into the brain's blood supply and vascular disease affecting small vessels. However, for quantitative characterization and precise representation of human angioarchitecture to, for example, inform blood-flow simulations, detailed segmentations of the smallest vessels are required. Given the success of deep learning-based methods in many segmentation tasks, we here explore their application to high-resolution MRA data, and address the difficulty of obtaining large data sets of correctly and comprehensively labelled data. We introduce VesselBoost, a vessel segmentation package, which utilizes deep learning and imperfect training labels for accurate vasculature segmentation. Combined with an innovative data augmentation technique, which leverages the resemblance of vascular structures, VesselBoost enables detailed vascular segmentations.

Joint B0 and Image Reconstruction in Low-Field MRI by Physics-Informed Deep-Learning.

OBJECTIVE: We present a model-based image reconstruction approach based on unrolled neural networks which corrects for image distortion and noise in low-field ( B0  âˆ¼  50mT) MRI. METHODS: Utilising knowledge about the underlying physics, a novel network architecture (SH-Net) is introduced which involves the estimation of spherical harmonic coefficients to guarantee a spatially smooth field map estimate. The SH-Net is integrated in an end-to-end trainable model which jointly estimates the B0-field map as well as the image. Experiments were conducted on retrospectively simulated low-field data of human knees. RESULTS: We compare our model to different model-based approaches at distinct noise levels and various B0-field distributions. Our results show that our physics-informed neural network approach outperforms the purely model-based methods by improving the PSNR up to 11.7% and the RMSE up to 86.3%. CONCLUSION: Our end-to-end trained model-based approach outperforms existing methods in reconstructing image and B0-field maps in the low-field regime. SIGNIFICANCE: low-field MRI is becoming increasingly more popular as it enables access to MR in challenging situations such as intensive care units or resource poor areas. Our method allows for fast and accurate image reconstruction in such low-field imaging with B0-inhomogeneity compensation under a wide range of various environmental conditions.

Cortical thickness of the posterior cingulate cortex is associated with the ketamine-induced altered sense of self: An ultra-high field MRI study.

Subanesthetic doses of ketamine induce an antidepressant effect within hours in individuals with treatment-resistant depression while it furthermore induces immediate but transient psychotomimetic effects. Among these psychotomimetic effects, an altered sense of self has specifically been associated with the antidepressant response to ketamine as well as psychedelics. However, there is plenty of variation in the extent of the drug-induced altered sense of self experience that might be explained by differences in basal morphological characteristics, such as cortical thickness. Regions that have been previously associated with a psychedelics-induced sense of self and with ketamine's mechanism of action, are the posterior cingulate cortex (PCC) and the pregenual anterior cingulate cortex (pgACC). In this randomized, placebo-controlled, double-blind cross-over magnetic resonance imaging study, thirty-five healthy male participants (mean age ± standard deviation (SD) = 25.1 ± 4.2 years) were scanned at 7 T. We investigated whether the cortical thickness of two DMN regions, the PCC and the pgACC, are associated with disembodiment and experience of unity scores, which were used to index the ketamine-induced altered sense of self. We observed a negative correlation between the PCC cortical thickness and the disembodiment scores (R = -0.54, p < 0.001). In contrast, no significant association was found between the pgACC cortical thickness and the ketamine-induced altered sense of self. In the context of the existing literature, our findings highlight the importance of the PCC as a structure involved in the mechanism of ketamine-induced altered sense of self that seems to be shared with different antidepressant agents with psychotomimetic effects operating on different classes of transmitter systems.

Exploration of Interpretability Techniques for Deep COVID-19 Classification Using Chest X-ray Images.

The outbreak of COVID-19 has shocked the entire world with its fairly rapid spread, and has challenged different sectors. One of the most effective ways to limit its spread is the early and accurate diagnosing of infected patients. Medical imaging, such as X-ray and computed tomography (CT), combined with the potential of artificial intelligence (AI), plays an essential role in supporting medical personnel in the diagnosis process. Thus, in this article, five different deep learning models (ResNet18, ResNet34, InceptionV3, InceptionResNetV2, and DenseNet161) and their ensemble, using majority voting, have been used to classify COVID-19, pneumoniæ and healthy subjects using chest X-ray images. Multilabel classification was performed to predict multiple pathologies for each patient, if present. Firstly, the interpretability of each of the networks was thoroughly studied using local interpretability methods-occlusion, saliency, input X gradient, guided backpropagation, integrated gradients, and DeepLIFT-and using a global technique-neuron activation profiles. The mean micro F1 score of the models for COVID-19 classifications ranged from 0.66 to 0.875, and was 0.89 for the ensemble of the network models. The qualitative results showed that the ResNets were the most interpretable models. This research demonstrates the importance of using interpretability methods to compare different models before making a decision regarding the best performing model.

Investigation of alternative RF power limit control methods for 0.5T, 1.5T, and 3T parallel transmission cardiac imaging: A simulation study.

PURPOSE: To investigate safety and performance aspects of parallel-transmit (pTx) RF control-modes for a body coil at B 0 ≤ 3 T $$ {B}_0\le 3\mathrm{T} $$ . METHODS: Electromagnetic simulations of 11 human voxel models in cardiac imaging position were conducted for B 0 = 0.5 T $$ {B}_0=0.5\mathrm{T} $$ , 1.5 T $$ 1.5\mathrm{T} $$ and 3 T $$ 3\mathrm{T} $$ and a body coil with a configurable number of transmit channels (1, 2, 4, 8, 16). Three safety modes were considered: the 'SAR-controlled mode' (SCM), where specific absorption rate (SAR) is limited directly, a 'phase agnostic SAR-controlled mode' (PASCM), where phase information is neglected, and a 'power-controlled mode' (PCM), where the voltage amplitude for each channel is limited. For either mode, safety limits were established based on a set of 'anchor' simulations and then evaluated in 'target' simulations on previously unseen models. The comparison allowed to derive safety factors accounting for varying patient anatomies. All control modes were compared in terms of the B 1 + $$ {B}_1^{+} $$ amplitude and homogeneity they permit under their respective safety requirements. RESULTS: Large safety factors (approximately five) are needed if only one or two anchor models are investigated but they shrink with increasing number of anchors. The achievable B 1 + $$ {B}_1^{+} $$ is highest for SCM but this advantage is reduced when the safety factor is included. PCM appears to be more robust against variations of subjects. PASCM performance is mostly in between SCM and PCM. Compared to standard circularly polarized (CP) excitation, pTx offers minor B 1 + $$ {B}_1^{+} $$ improvements if local SAR limits are always enforced. CONCLUSION: PTx body coils can safely be used at B 0 ≤ 3 T $$ {B}_0\le 3\mathrm{T} $$ . Uncertainties in patient anatomy must be accounted for, however, by simulating many models.

Reduced dimension stimulus decoding and column-based modeling reveal architectural differences of primary somatosensory finger maps between younger and older adults.

The primary somatosensory cortex (SI) contains fine-grained tactile representations of the body, arranged in an orderly fashion. The use of ultra-high resolution fMRI data to detect group differences, for example between younger and older adults' SI maps, is challenging, because group alignment often does not preserve the high spatial detail of the data. Here, we use robust-shared response modeling (rSRM) that allows group analyses by mapping individual stimulus-driven responses to a lower dimensional shared feature space, to detect age-related differences in tactile representations between younger and older adults using 7T-fMRI data. Using this method, we show that finger representations are more precise in Brodmann-Area (BA) 3b and BA1 compared to BA2 and motor areas, and that this hierarchical processing is preserved across age groups. By combining rSRM with column-based decoding (C-SRM), we further show that the number of columns that optimally describes finger maps in SI is higher in younger compared to older adults in BA1, indicating a greater columnar size in older adults' SI. Taken together, we conclude that rSRM is suitable for finding fine-grained group differences in ultra-high resolution fMRI data, and we provide first evidence that the columnar architecture in SI changes with increasing age.

Sinogram upsampling using Primal-Dual UNet for undersampled CT and radial MRI reconstruction.

Computed tomography (CT) and magnetic resonance imaging (MRI) are two widely used clinical imaging modalities for non-invasive diagnosis. However, both of these modalities come with certain problems. CT uses harmful ionising radiation, and MRI suffers from slow acquisition speed. Both problems can be tackled by undersampling, such as sparse sampling. However, such undersampled data leads to lower resolution and introduces artefacts. Several techniques, including deep learning based methods, have been proposed to reconstruct such data. However, the undersampled reconstruction problem for these two modalities was always considered as two different problems and tackled separately by different research works. This paper proposes a unified solution for both sparse CT and undersampled radial MRI reconstruction, achieved by applying Fourier transform-based pre-processing on the radial MRI and then finally reconstructing both modalities using sinogram upsampling combined with filtered back-projection. The Primal-Dual network is a deep learning based method for reconstructing sparsely-sampled CT data. This paper introduces Primal-Dual UNet, which improves the Primal-Dual network in terms of accuracy and reconstruction speed. The proposed method resulted in an average SSIM of 0.932±0.021 while performing sparse CT reconstruction for fan-beam geometry with a sparsity level of 16, achieving a statistically significant improvement over the previous model, which resulted in 0.919±0.016. Furthermore, the proposed model resulted in 0.903±0.019 and 0.957±0.023 average SSIM while reconstructing undersampled brain and abdominal MRI data with an acceleration factor of 16, respectively - statistically significant improvements over the original model, which resulted in 0.867±0.025 and 0.949±0.025. Finally, this paper shows that the proposed network not only improves the overall image quality, but also improves the image quality for the regions-of-interest: liver, kidneys, and spleen; as well as generalises better than the baselines in presence the of a needle.

MICDIR: Multi-scale inverse-consistent deformable image registration using UNetMSS with self-constructing graph latent.

Image registration is the process of bringing different images into a common coordinate system - a technique widely used in various applications of computer vision, such as remote sensing, image retrieval, and, most commonly, medical imaging. Deep learning based techniques have been applied successfully to tackle various complex medical image processing problems, including medical image registration. Over the years, several image registration techniques have been proposed using deep learning. Deformable image registration techniques such as Voxelmorph have been successful in capturing finer changes and providing smoother deformations. However, Voxelmorph, as well as ICNet and FIRE, do not explicitly encode global dependencies (i.e. the overall anatomical view of the supplied image) and, therefore, cannot track large deformations. In order to tackle the aforementioned problems, this paper extends the Voxelmorph approach in three different ways. To improve the performance in case of small as well as large deformations, supervision of the model at different resolutions has been integrated using a multi-scale UNet. To support the network to learn and encode the minute structural co-relations of the given image-pairs, a self-constructing graph network (SCGNet) has been used as the latent of the multi-scale UNet - which can improve the learning process of the model and help the model to generalise better. And finally, to make the deformations inverse-consistent, cycle consistency loss has been employed. On the task of registration of brain MRIs, the proposed method achieved significant improvements over ANTs and VoxelMorph, obtaining a Dice score of 0.8013 ± 0.0243 for intramodal and 0.6211 ± 0.0309 for intermodal, while VoxelMorph achieved 0.7747 ± 0.0260 and 0.6071 ± 0.0510, respectively.

Cross-sectional study of retrospective self-reported childhood emotional neglect and inhibitory neurometabolite levels in the pregenual anterior cingulate cortex in adult humans.

High childhood emotional maltreatment (CM-EMO) is reported in mood and anxiety disorders. The associations with an increased risk for psychopathology are not fully understood. One potential factor may be through alterations in gamma-Aminobutyric acid (GABA). The pregenual anterior cingulate cortex (pgACC) is an important brain region for emotion processing and its' GABA levels were previously implicated in mood and anxiety disorders pathophysiology. We examined the association between the self-reported CM-EMO in adulthood and GABA + levels in the pgACC and in a control region, anterior mid cingulate cortex. GABA+ and total creatine (tCr) were measured in the pgACC and aMCC voxels in seventy-four healthy volunteers (32 (43%) women, ages 19-54, age [standard deviation] = 27.1 [6.5]) using proton magnetic resonance spectroscopy at 7 T. Childhood Trauma Questionnaire was completed by adult participants to measure retrospective self-reported experience of emotional neglect (CM-EMO-NEG) and emotional abuse (CM-EMO-AB) during childhood. Linear mixed models tested the interaction between the region and the two subscales, and GABA+/tCr ratios, with an adjusted alpha = 0.025. Following, linear models, including with covariates were tested. There was an interaction effect between region and CM-EMO-NEG (B = -0.007, p = 0.009), driven by a negative relationship between CM-EMO-NEG and GABA+/tCr in the pgACC (B = -0.004, p = 0.013). Results for CM-EMO-NEG were robust to inclusion of different covariates (ps < 0.035). There was no interaction effect for the CM-EMO-AB (B = 0.007, p = 0.4). Limitations include cross-sectional measurement and retrospective nature of the CTQ. The findings indicate preliminary importance of inhibitory neurometabolite concentrations in the pgACC for retrospective reporting of CM-EMO-NEG.

It is the locus coeruleus! Or is it?: a proposition for analyses and reporting standards for structural and functional magnetic resonance imaging of the noradrenergic locus coeruleus.

The noradrenergic locus coeruleus (LC) is one of the protein pathology epicenters in neurodegenerative diseases. In contrast to PET (positron emission tomography), MRI (magnetic resonance imaging) offers the spatial resolution necessary to investigate the 3-4 mm wide and 1.5 cm long LC. However, standard data postprocessing is often too spatially imprecise to allow investigating the structure and function of the LC at the group level. Our analysis pipeline uses a combination of existing toolboxes (SPM12, ANTs, FSL, FreeSurfer), and is tailored towards achieving suitable spatial precision in the brainstem area. Its effectiveness is demonstrated using 2 datasets comprising both younger and older adults. We also suggest quality assessment procedures which allow to quantify the spatial precision obtained. Spatial deviations below 2.5 mm in the LC area are achieved, which is superior to current standard approaches. Relevant for ageing and clinical researchers interested in brainstem imaging, we provide a tool for more reliable analyses of structural and functional LC imaging data which can be also adapted for investigating other nuclei of the brainstem.

Cognitive training, but not EEG-neurofeedback, improves working memory in healthy volunteers.

Working memory performance can be influenced by motivational factors, which may be associated with specific brain activities, including suppression of alpha oscillations. We investigated whether providing individuals online feedback about their ongoing oscillations (EEG-neurofeedback) can improve working memory under high and low reward expectancies. We combined working memory training with neurofeedback to enhance alpha suppression in a monetary-rewarded delayed match-to-sample task for visual objects. Along with alpha, we considered the neighbouring theta and beta bands. In a double-blind experiment, individuals were trained over 5 days to suppress alpha power by receiving real-time neurofeedback or control neurofeedback (placebo) in reward and no-reward trials. We investigated (i) whether neurofeedback enhances alpha suppression, (ii) whether monetary reward enhances alpha suppression and working memory, and (iii) whether any performance benefits of neurofeedback-training would transfer to unrelated cognitive tasks. With the same experimental design, we conducted two studies with differing instructions given at the maintenance, yielding together 300 EEG recording sessions. In Study I, participants were engaged in a mental calculation task during maintenance. In Study II, they were instructed to visually rehearse the sample image. Results from Study I demonstrated a significant training and reward-anticipation effect on working memory accuracy and reaction times over 5 days. Neurofeedback and reward anticipation showed effects on theta suppression but not on alpha suppression. Moreover, a cognitive training effect was observed on beta suppression. Thus, neurofeedback-training of alpha was unrelated to working memory performance. Study II replicated the training and reward-anticipation effect on working memory but without any effects of neurofeedback-training on oscillations or working memory. Neither study showed transfer effects of either working memory or neurofeedback-training. A linear mixed-effect model analysis of neurofeedback-independent training-related improvement of working memory combining both studies showed that improved working memory performance was related to oscillatory changes over training days in the encoding and maintenance phases. Improvements in accuracy were related to increasing beta amplitude in reward trials over right parietal electrodes. Improvements in reaction times were related to increases in right parietal theta amplitude during encoding and increased right parietal and decreased left parietal beta amplitudes during maintenance. Thus, while our study provided no evidence that neurofeedback targeting alpha improved the efficacy of working memory training or evidence for transfer, it showed a relationship between training-related changes in parietal beta oscillations during encoding and improvements in accuracy. Right parietal beta oscillations could be an intervention target for improving working memory accuracy.

Germany's journey toward 14 Tesla human magnetic resonance.

Multiple sites within Germany operate human MRI systems with magnetic fields either at 7 Tesla or 9.4 Tesla. In 2013, these sites formed a network to facilitate and harmonize the research being conducted at the different sites and make this technology available to a larger community of researchers and clinicians not only within Germany, but also worldwide. The German Ultrahigh Field Imaging (GUFI) network has defined a strategic goal to establish a 14 Tesla whole-body human MRI system as a national research resource in Germany as the next progression in magnetic field strength. This paper summarizes the history of this initiative, the current status, the motivation for pursuing MR imaging and spectroscopy at such a high magnetic field strength, and the technical and funding challenges involved. It focuses on the scientific and science policy process from the perspective in Germany, and is not intended to be a comprehensive systematic review of the benefits and technical challenges of higher field strengths.

Vessel distance mapping: A novel methodology for assessing vascular-induced cognitive resilience.

The association between cerebral blood supply and cognition has been widely discussed in the recent literature. One focus of this discussion has been the anatomical variability of the circle of Willis, with morphological differences being present in more than half of the general population. While previous studies have attempted to classify these differences and explore their contribution to hippocampal blood supply and cognition, results have been controversial. To disentangle these previously inconsistent findings, we introduce Vessel Distance Mapping (VDM) as a novel methodology for evaluating blood supply, which allows for obtaining vessel pattern metrics with respect to the surrounding structures, extending the previously established binary classification into a continuous spectrum. To accomplish this, we manually segmented hippocampal vessels obtained from high-resolution 7T time-of-flight MR angiographic imaging in older adults with and without cerebral small vessel disease, generating vessel distance maps by computing the distances of each voxel to its nearest vessel. Greater values of VDM-metrics, which reflected higher vessel distances, were associated with poorer cognitive outcomes in subjects affected by vascular pathology, while this relation was not observed in healthy controls. Therefore, a mixed contribution of vessel pattern and vessel density is proposed to confer cognitive resilience, consistent with previous research findings. In conclusion, VDM provides a novel platform, based on a statistically robust and quantitative method of vascular mapping, for addressing a variety of clinical research questions.

The 3D Structural Architecture of the Human Hand Area Is Nontopographic.

The functional topography of the human primary somatosensory cortex hand area is a widely studied model system to understand sensory organization and plasticity. It is so far unclear whether the underlying 3D structural architecture also shows a topographic organization. We used 7 Tesla (7T) magnetic resonance imaging (MRI) data to quantify layer-specific myelin, iron, and mineralization in relation to population receptive field maps of individual finger representations in Brodman area 3b (BA 3b) of human S1 in female and male younger adults. This 3D description allowed us to identify a characteristic profile of layer-specific myelin and iron deposition in the BA 3b hand area, but revealed an absence of structural differences, an absence of low-myelin borders, and high similarity of 3D microstructure profiles between individual fingers. However, structural differences and borders were detected between the hand and face areas. We conclude that the 3D structural architecture of the human hand area is nontopographic, unlike in some monkey species, which suggests a high degree of flexibility for functional finger organization and a new perspective on human topographic plasticity.SIGNIFICANCE STATEMENT Using ultra-high-field MRI, we provide the first comprehensive in vivo description of the 3D structural architecture of the human BA 3b hand area in relation to functional population receptive field maps. High similarity of precise finger-specific 3D profiles, together with an absence of structural differences and an absence of low-myelin borders between individual fingers, reveals the 3D structural architecture of the human hand area to be nontopographic. This suggests reduced structural limitations to cortical plasticity and reorganization and allows for shared representational features across fingers.

Association of the delayed changes in glutamate levels and functional connectivity with the immediate network effects of S-ketamine.

Ketamine shows rapid antidepressant effects peaking 24 h after administration. The antidepressant effects may occur through changes in glutamatergic metabolite levels and resting-state functional connectivity (rsFC) within the default mode network (DMN). A multistage drug effect of ketamine has been suggested, inducing acute effects on dysfunctional network configuration and delayed effects on homeostatic synaptic plasticity. Whether the DMN-centered delayed antidepressant-related changes are associated with the immediate changes remains unknown. Thirty-five healthy male participants (25.1 ± 4.2 years) underwent 7 T magnetic resonance spectroscopy (MRS) and resting-state functional magnetic resonance imaging (rsfMRI) before, during, and 24 h after a single S-ketamine or placebo infusion. Changes in glutamatergic measures and rsFC in the DMN node pregenual anterior cingulate cortex (pgACC) were examined. A delayed rsFC decrease of the pgACC to inferior parietal lobe (family-wise error corrected p (pFWEc) = 0.018) and dorsolateral prefrontal cortex (PFC; pFWEc = 0.002) was detected that was preceded by an immediate rsFC increase of the pgACC to medial PFC (pFWEc < 0.001) and dorsomedial PFC (pFWEc = 0.005). Additionally, the immediate rsFC reconfigurations correlated with the delayed pgACC glutamate (Glu) level increase (p = 0.024) after 24 h at trend level (p = 0.067). Baseline measures of rsFC and MRS were furthermore associated with the magnitude of the respective delayed changes (p's < 0.05). In contrast, the delayed changes were not associated with acute psychotomimetic side effects or plasma concentrations of ketamine and its metabolites. This multimodal study suggests an association between immediate S-ketamine-induced network effects and delayed brain changes at a time point relevant in its clinical context.

Towards an integrated radiofrequency safety concept for implant carriers in MRI based on sensor-equipped implants and parallel transmission.

To protect implant carriers in MRI from excessive radiofrequency (RF) heating it has previously been suggested to assess that hazard via sensors on the implant. Other work recommended parallel transmission (pTx) to actively mitigate implant-related heating. Here, both ideas are integrated into one comprehensive safety concept where native pTx safety (without implant) is ensured by state-of-the-art field simulations and the implant-specific hazard is quantified in situ using physical sensors. The concept is demonstrated by electromagnetic simulations performed on a human voxel model with a simplified spinal-cord implant in an eight-channel pTx body coil at 3 T . To integrate implant and native safety, the sensor signal must be calibrated in terms of an established safety metric (e.g., specific absorption rate [SAR]). Virtual experiments show that E -field and implant-current sensors are well suited for this purpose, while temperature sensors require some caution, and B 1 probes are inadequate. Based on an implant sensor matrix Q s , constructed in situ from sensor readings, and precomputed native SAR limits, a vector space of safe RF excitations is determined where both global (native) and local (implant-related) safety requirements are satisfied. Within this safe-excitation subspace, the solution with the best image quality in terms of B 1 + magnitude and homogeneity is then found by a straightforward optimization algorithm. In the investigated example, the optimized pTx shim provides a 3-fold higher mean B 1 + magnitude compared with circularly polarized excitation for a maximum implant-related temperature increase ∆ T imp ≤ 1 K . To date, sensor-equipped implants interfaced to a pTx scanner exist as demonstrator items in research labs, but commercial devices are not yet within sight. This paper aims to demonstrate the significant benefits of such an approach and how this could impact implant-related RF safety in MRI. Today, the responsibility for safe implant scanning lies with the implant manufacturer and the MRI operator; within the sensor concept, the MRI manufacturer would assume much of the operator's current responsibility.

Liver segmentation using Turbolift learning for CT and cone-beam C-arm perfusion imaging.

Model-based reconstruction employing the time separation technique (TST) was found to improve dynamic perfusion imaging of the liver using C-arm cone-beam computed tomography (CBCT). To apply TST using prior knowledge extracted from CT perfusion data, the liver should be accurately segmented from the CT scans. Reconstructions of primary and model-based CBCT data need to be segmented for proper visualisation and interpretation of perfusion maps. This research proposes Turbolift learning, which trains a modified version of the multi-scale Attention UNet on different liver segmentation tasks serially, following the order of the trainings CT, CBCT, CBCT TST - making the previous trainings act as pre-training stages for the subsequent ones - addressing the problem of limited number of datasets for training. For the final task of liver segmentation from CBCT TST, the proposed method achieved an overall Dice scores of 0.874±0.031 and 0.905±0.007 in 6-fold and 4-fold cross-validation experiments, respectively - securing statistically significant improvements over the model, which was trained only for that task. Experiments revealed that Turbolift not only improves the overall performance of the model but also makes it robust against artefacts originating from the embolisation materials and truncation artefacts. Additionally, in-depth analyses confirmed the order of the segmentation tasks. This paper shows the potential of segmenting the liver from CT, CBCT, and CBCT TST, learning from the available limited training data, which can possibly be used in the future for the visualisation and evaluation of the perfusion maps for the treatment evaluation of liver diseases.