Publications

{Functional imaging with sub-millimeter spatial resolution is a basic requirement for assessing functional MRI (fMRI) responses across different cortical depths, and is used extensively in the emerging field of laminar fMRI. Such studies seek to investigate the detailed functional organization of the brain and may develop to a new powerful tool for human neuroscience. However, several studies have shown that measurement of laminar fMRI responses can be biased by the image acquisition and data processing strategies. In this work, measurements with three different gradient-echo EPI protocols with a voxel size down to 650 $\mu$m isotropic were performed at 9.4 T. We estimated how prospective motion correction can help to improve spatial accuracy by reducing the number of spatial resampling steps in postprocessing. In addition, we demonstrate key requirements for accurate geometric distortion correction to ensure that distortion correction maps are properly aligned to the functional data and that strong variations of distortions near large veins can lead to signal overlays which cannot be corrected for during postprocessing. Furthermore, this study illustrates the spatial extent of bias induced by pial and other larger veins in laminar BOLD experiments. Since these issues under investigation affect studies performed with more conventional spatial resolutions, the methods applied in this work may also help to improve the understanding of the BOLD signal more broadly.}

{Cocaine addiction is characterized by overwhelming craving for the substance, which drives its escalating use despite adverse consequences. Animal models suggest a disrupted glutamate homeostasis in the nucleus accumbens to underlie addiction-like behavior. After chronic administration of cocaine, rodents show decreased levels of accumbal glutamate, whereas drug-seeking reinstatement is associated with enhanced glutamatergic transmission. However, due to technical obstacles, the role of disturbed glutamate homeostasis for cocaine addiction in humans remains only partially understood, and accordingly, no approved pharmacotherapy exists. Here, we applied a tailored proton magnetic resonance spectroscopy protocol that allows glutamate quantification within the human nucleus accumbens. We found significantly reduced basal glutamate concentrations in the nucleus accumbens in cocaine-addicted (N \textequals 26) compared with healthy individuals (N \textequals 30), and increased glutamate levels during cue-induced craving in cocaine-addicted individuals compared with baseline. These glutamatergic alterations, however, could not be significantly modulated by a short-term challenge of N-acetylcysteine (2400 mg/day on 2 days). Taken together, our findings reveal a disturbed accumbal glutamate homeostasis as a key neurometabolic feature of cocaine addiction also in humans. Therefore, we suggest the glutamatergic system as a promising target for the development of novel pharmacotherapies, and in addition, as a potential biomarker for a personalized medicine approach in addiction.}

{Using a contingency volatility manipulation, we tested the hypothesis that difficulty adapting probabilistic decision-making to second-order uncertainty might reflect a core deficit that cuts across anxiety and depression and holds regardless of whether outcomes are aversive or involve reward gain or loss. We used bifactor modeling of internalizing symptoms to separate symptom variance common to both anxiety and depression from that unique to each. Across two experiments, we modeled performance on a probabilistic decision-making under volatility task using a hierarchical Bayesian framework. Elevated scores on the common internalizing factor, with high loadings across anxiety and depression items, were linked to impoverished adjustment of learning to volatility regardless of whether outcomes involved reward gain, electrical stimulation, or reward loss. In particular, high common factor scores were linked to dampened learning following better-than-expected outcomes in volatile environments. No such relationships were observed for anxiety- or depression-specific symptom factors.}

{It has been shown that neural networks combined with variable k-space undersampling (MultiNet GRAPPA) is superior to a conventional GRAPPA reconstruction at 9.4T. Here, the feasibility of performing MultiNet GRAPPA for 1H FID-MRSI at 7T is investigated with and without novel modifications to the original acquisition/reconstruction scheme. In this study, it is shown that MultiNet GRAPPA is shown to be feasible for 1H MRSI acceleration at 7T with a new k-space undersampling scheme for higher signal-to-noise and increased map reliability and use of a novel technique to increase SNR retention using semi-synthetic calibration data without an increase in acquisition time.}

{The double-echo steady-state (DESS) sequence has been used successfully in 3T MRI imaging of the musculoskeletal system for segmentation of joint. However, in 3D-DESS images acquired at 7T MRI, a reduction in the contrast between tissues due to an increased diffusion sensitivity may complicate cartilage segmentation. Typically, the signals acquired with DESS are averaged without any pre-processing. However, these signals give different contrasts and have different noise behaviours. In this work, we improve the contrast-to-noise ratio (CNR) whilst preserving anatomical detail in high-resolution 7T DESS images through a new approach combining L1-norm denoising and p-norm combination of the echo signals.}

{We report on a macrocyclic platform based on an 18-membered macrocycle that forms kinetically highly inert paramagnetic complexes and possesses an excellent outlook for the development of bioresponsive paraCEST (paramagnetic chemical exchange saturation transfer) contrast agents. The investigated europium(III) chelate is non-hydrated and contains four amide groups, each possessing two paramagnetically shifted proton resonances distant from bulk water. The X-ray crystal structure and solution studies indicate that the metal ion is ten-coordinated, being directly bound to the six N atoms of the macrocycle and the four amide O atoms of the pendant arms. The complex presents an excellent inertness with respect to dissociation, being stable under a variety of harsh conditions, including highly acidic and basic media or elevated temperatures. The amide protons are in slow-to-intermediate exchange with bulk water, which gives rise to the generation of a strong CEST effect at low probe concentration and saturation powers ($\sim$25\textpercent at 5 mM, B1 \textequals 5 $\mu$T, 37 \mbox{$^\circ$}C). We demonstrate the potential of this platform for mapping pH in its microenvironment and foresee potential for the development of diverse paraCEST probes and sensors.}

{Introduction: In functional MRI (fMRI) echo planar imaging (EPI) is often combined with parallel imaging, e.g. GRAPPA (1), to increase temporal resolution. The auto-calibration scans (ACS) required for the calculation of the coil sensitivities in the parallel imaging reconstruction are conventionally acquired in a segmented fashion (number of segments \textequals parallel imaging factor), with the individual segments of each slice separated by the repetition time (TR). However, in case of TRs in the range of several seconds, ACS segments may be acquired at different B0-field offsets e.g. due to respiration or motion. These fluctuations can result in variations in temporal SNR (tSNR) across different slices particularly at high-field (3). The sensitivity of tSNR on physiological effects can be reduced by acquiring all segments of a slice successively with minimum delay in the so called FLEET technique (3). Alternatively, a FLASH readout, which is more robust against B0-field changes, can be used to obtain the ACS data (2). Although physiological influences are usually considered to be the main cause of tSNR variations at long TRs, as far as we know, the performance of various GRAPPA pre-scan methods (conventional, FLEET and FLASH) has not previously been investigated for a TR in the sub-second range. Methods: Four healthy subjects were measured at 9.4 Tesla (Siemens Healthineers, Germany) using an in-house-built 16Tx-31Rx head-coil (4). Gradient-echo EPIs were acquired for two regions covering a major part of the thalamus (ROI 1) and the motor cortex (ROI 2). Imaging parameters: TE/TR \textequals 23/600ms, FA \textequals 40\mbox{$^\circ$}, 12 slices, 150 volumes. Two different spatial resolutions were used: \textbullet 1 x 1 x 2 mm\mbox{$^3$}: mtx \textequals 192x192, 6/8 partial Fourier, GRAPPA \textequals 4 (60 ACS lines), echo spacing \textequals 1.01 ms. \textbullet 2 x 2 x 2 mm\mbox{$^3$}: mtx \textequals 96x96, GRAPPA \textequals 3 (45 ACS lines), echo spacing \textequals 0.8 ms. The two protocols were repeated for both ROIs for all three ACS sampling methods: conventional, FLEET, and FLASH. The excitation flip angle for the FLEET and FLASH ACS scans was 10\mbox{$^\circ$} and 15\mbox{$^\circ$}, respectively. Temporal SNR maps were calculated as the mean signal value across time divided by its temporal standard deviation. To quantify the tSNR for the different GRAPPA pre-scan methods, mean tSNR values were assessed for each ROI after performing manual brain masking. Results: Figure 1 shows the calculated tSNR for the different GRAPPA pre-scan methods and brain regions in an example volunteer. The lowest tSNR is visible for the data measured with conventional ACS and low spatial resolution in particular. This observation is consistent for both ROIs. Averaged over all slices, the tSNR values in images acquired with FLEET or FLASH ACS sampling are higher than with conventional ordering, too (Figure 2). This is especially the case at low image resolution. At high spatial resolution, the tSNR of data reconstructed using FLEET and FLASH sampled data is almost identical and the improvement compared to the conventional method is rather small ($\sim$12\textpercent in ROI 1 and $\sim$25\textpercent in ROI 2). Conclusions: Although physiological influences and respiration effects in particular are expected to be reduced for sub-second TR, the FLEET and FLASH pre-scan methods yielded clearly higher tSNR compared to the conventional approach. One explanation is, that despite the short TR, the acquisition of all ACS lines still took about 1.8 s (2x2x2 mm\mbox{$^3$}) and 2.4 s (1x1x2 mm\mbox{$^3$}), respectively, due to the slice-segment acquisition scheme, whereas the FLEET method only required about 200 ms (1x1x2 mm\mbox{$^3$}). This study also confirmed that the impact of physiological fluctuations on tSNR heavily scales with the spatial resolution, as it is the case for un-accelerated imaging (5). Thus, even though less tSNR improvement can be expected for alternative ACS acquisition techniques at high spatial resolutions, it still has to be considered as a potential source for effect size differences even in sub-second TR fMRI studies.}

{Post-mortem brain MRI can yield valuable information. However, tissue preservation and MR-compatibility of fixation agents are challenging. Here we investigated the effect of four MR-compatible formalin-based fixatives on the MR properties of pig brains at several timepoints after start of fixation up to one month. The inclusion agents known to improve the dielectric properties of the fixatives lead to greater R2\textasteriskcentered difference between GM-WM than conventional fixatives. Vice versa these agents lead to a decrease in T1 contrast.}

{The brain has persistent internal states that can modulate every aspect of an animal\textquoterights mental experience1,2,3,4. In complex tasks such as foraging, the internal state is dynamic5,6,7,8. Caenorhabditis elegans alternate between local search and global dispersal5. Rodents and primates exhibit trade-offs between exploitation and exploration6,7. However, fundamental questions remain about how persistent states are maintained in the brain, which upstream networks drive state transitions and how state-encoding neurons exert neuromodulatory effects on sensory perception and decision-making to govern appropriate behaviour. Here, using tracking microscopy to monitor whole-brain neuronal activity at cellular resolution in freely moving zebrafish larvae9, we show that zebrafish spontaneously alternate between two persistent internal states during foraging for live prey (Paramecia). In the exploitation state, the animal inhibits locomotion and promotes hunting, generating small, localized trajectories. In the exploration state, the animal promotes locomotion and suppresses hunting, generating long-ranging trajectories that enhance spatial dispersion. We uncover a dorsal raphe subpopulation with persistent activity that robustly encodes the exploitation state. The exploitation-state-encoding neurons, together with a multimodal trigger network that is associated with state transitions, form a stochastically activated nonlinear dynamical system. The activity of this oscillatory network correlates with a global retuning of sensorimotor transformations during foraging that leads to marked changes in both the motivation to hunt for prey and the accuracy of motor sequences during hunting. This work reveals an important hidden variable that shapes the temporal structure of motivation and decision-making.}

{There is growing evidence that obesity is associated with inflammation in the brain, which could contribute to the pathogenesis of obesity. In humans, it is challenging to detect brain inflammation in vivo. Recently, quantitative magnetic resonance imaging (qMRI) has emerged as a tool to characterize pathophysiological processes in the brain with reliable and reproducible measures. Proton density imaging provides quantitative assessment of the brain water content, which is affected in different pathologies including inflammation. We enrolled 115 normal weight, overweight and obese men and women (body mass index (BMI) range 20.1-39.7 kg/m2, age range 20-75 years, 60\textpercent men) to acquire cerebral water content mapping in vivo using MRI at 3 Tesla. We investigated potential associations between brain water content with anthropometric measures of obesity, body fat distribution and whole-body metabolism. No global changes in water content were associated with obesity. However, higher water content values in the cerebellum, limbic lobe and sub-lobular region was detected in participants with higher BMI, independent of age. More specifically, the dorsal striatum, hypothalamus, thalamus, fornix, anterior limb of the internal capsule and posterior thalamic radiation showed the strongest relationship with BMI, independent of age. In a subgroup with available measurements (n\textequals50), we identified visceral adipose tissue to be the strongest tested link between higher water content values and obesity. Persons with metabolic syndrome had the highest water content values in the hypothalamus and the fornix. There is accumulating evidence that inflammation of the hypothalamus contributed to obesity-associated insulin resistance in that area. Whether brain inflammation is a cause or consequence of obesity in humans still needs to be investigated using a longitudinal study design. Using qMRI, we were able to detect marked water content changes in young and older obese adults, which is most likely due to chronic low-grade inflammation.}

{Many processes in the brain require a constant supply of energy. Notably, spiking and maintenance of ion concentration gradients by the Na/K pump require active mechanisms and consumes energy in the form of ATP. Though the brain has a safety margin on energy production compared to the consumption during maximum activity, this margin is small and energy stores can be quickly exhausted, for instance in the context of neuronal disorders [1]. Once energy becomes insufficient, neuronal response can change drastically, leading to intermittent spikes, bursts, network oscillations, or seizures [2]. This crucial impact of metabolic disruption is not addressed in any of the standard neuronal models used in computational neuroscience. We fill this significant gap in current modeling frameworks, by introducing two novel model-neurons that account for energetic constraints while remaining computationally efficient, analytically tractable, and biologically interpretable. Using these new types of neurons, we can reproduce both crucial single-cell behaviors, such as depolarization blocks or bistability, and complex changes in collective dynamics that were not capture by previous integrate-and-fire neurons (shown on panels A and B of the figure). In this presentation, we will discuss the main mechanisms that define the models\textquoteright equations: the role of pumps that degrade ATP into ADP to maintain ion gradients (most notably the Na/K and calcium pumps) and ATP-gated potassium (K-ATP) channels, which open or close depending on the ATP/ADP ratio. We explain how these mechanisms shape the relationship between energy availability and neuronal excitability [3-4] and illustrate some of the potential consequences. In particular we discuss the influence of metabolic disruption on the information processing capabilities of neurons (see panel A for an example of disease progression on a network in the well-known AI state). These novel models enable for the first time the theoretical study of energy-associated disorders over the whole time-course of disease progression, instead of only comparing the initially healthy condition with the final diseased state. This advancement is essential to model multiple neurological disorders, such as epilepsy or Parkinson\textquoterights disease, as it enables theoretical and computational studies to assess the opportunities of early diagnostics and the potential of energy-centered approaches to improve therapies.}

{In the following report, using a model adapted to real-world data of the spread of COVID-19, it is shown that if COVID-19 is spread in Iran without any containment policy, which is more or less the case at the moment of writing this text, it can lead to disastrous consequences. Although the reports show that the current situation in Iran is already very bad, simulations show that the worse is yet to come. The peak number of infected people in Iran will probably happen in 1.5-2 months from now. Simulations show that in an uncontained human population with the age structure of the Iranian population, at the peak of the infection, around 11\textpercent of the population will be infected simultaneously. And eventually, 65\textpercent of the population will be infected. For a country as big as Iran, these peak values might reach in different days in different cities or regions. But all are expected to reach these peak values days or weeks apart. If we assume only 5\textpercent of those infected with COVID-19 need respiratory or intensive care, the total number of hospital beds needed at the peak of the epidemic in Iran is around 467, 000 beds. Also, in the uncontained scenario, eventually 65\textpercent of the population will be infected with COVID-19 in a matter of months, which is around 55 million people for a country of around 85 million inhabitants. Even with a mortality rate of 0.2\textpercent, which admittedly is optimistic for such a huge number of potential patients, total death would amount to around 110, 000 people. To avoid such an unprecedented disaster, there is an immediate need to implement effective containment strategies. I have presented a detailed quantitative analysis of the effectiveness of various containment policies. The most effective, among the ones humanly possible, is a total lock-down of the population, closing down all governmental offices, companies, banks and schools and forcing people to stay home. A policy which is already enforced by governments in different countries around the world. Simulations show that if such a policy is imposed today, it can bring down the total number of infected to around 7\textpercent of the population, down from 65\textpercent. But even in such a case, considering the current level of the spread of the COVID-19 virus in the Iranian population, eradicating it would take months. It needs patience and determination and a long-term strategy to manage the number of infected people in the population until the vaccine for COVID-19 is publicly available.}

{In this paper, we investigate the principle that \textasciigravegood explanations are hard to vary\textquotesingle in the context of deep learning. We show that averaging gradients across examples -- akin to a logical OR of patterns -- can favor memorization and \textasciigravepatchwork\textquotesingle solutions that sew together different strategies, instead of identifying invariances. To inspect this, we first formalize a notion of consistency for minima of the loss surface, which measures to what extent a minimum appears only when examples are pooled. We then propose and experimentally validate a simple alternative algorithm based on a logical AND, that focuses on invariances and prevents memorization in a set of real-world tasks. Finally, using a synthetic dataset with a clear distinction between invariant and spurious mechanisms, we dissect learning signals and compare this approach to well-established regularizers.}

{Why we learn, what we learn, what the conditions are under which we learn, how the learning happens, and what the answers to these questions imply about how we should teach are the topics of this urgent, compelling, and highly multidisciplinary book by Stanislas Dehaene. He is one of the world\textquoterights most storied and celebrated cognitive scientists and cognitive neuroscientists. Along the way, we get to enjoy some mental gymnastics concerning the joint roles of nature and nurture, a wealth of revealing and sometimes heart-wrenching individual examples, and a glimpse into the potential potency but current impoverishment of machine learning systems.}

{How do people learn complex rules? We introduce a novel paradigm called \textquotedblrightTrack-A-Mole\textquotedblright, in which participants have to learn about and predict the moves of a cartoon mole, whose movements are generated by graphical programs. Our results show that participants can learn to predict richly structured programs, and often require only few observations to do so, showing rapid learning and early insights about the underlying patterns. Moreover, we found that how learnable a program is can be predicted by features related to its complexity and compressibility. Finally, participants also show interesting patterns of generalizations, assuming more parsimonious rules first and then gradually adjusting their predictions to more complex regularities, as well as matching their predictions to the general direction of movements and producing sensi- ble errors. These results extend our understanding of complex rule learning and open up future opportunities to model sequential pattern predictions as graphical program induction.}

{As we transition from child to adult, we navigate the worlddifferently. In this world, many of the relationships betweenevents are unclear or uncertain because they are probabilisticin nature. We wanted to know how learning about probabilis-tic relationships changes with development and to interrogatethe underlying processes. We investigated these questions in aprobabilistic reinforcement learning task (The Butterfly Task)with 302 participants aged 8-30. We found performance in thistask increased with age through early-twenties, then stabilized.Using hierarchical Bayesian methods to fit computational rein-forcement learning models, we showed that this performanceincrease was driven by 1) an increase in learning rate (i.e. de-crease in integration time horizon); 2) a decrease in exploratorychoices. By contrast, forgetting rates did not change with age.We discuss our findings in the context of other studies and hy-potheses about adolescent brain development.}

{Purpose: To validate the feasibility of localized B0 shimming based on B0 maps acquired with sodium (23Na) MRI. Methods A localized B0 shimming routine based on a constrained regularized algorithm in combination with 23Na MRI data acquired with a 3D density-adapted radial readout scheme was implemented on a 7T MR system. Measurements were performed using a dual-tuned 23Na/1H head coil. The quality of B0 maps reconstructed from 23Na images and the resulting shim values was examined depending on the acquisition duration between 10 minutes and 15 seconds to examine clinical applicability. The B0 shimming based on 23Na B0 maps was performed both for phantom and human head of 6 healthy volunteers, and the resulting B0 homogeneity was compared with the vendor-provided 1H MRI\textendashbased gradient-echo brain shimming routine. Results The proposed 23Na MRI\textendashbased shimming routine showed a reduction in B0 variation comparable to the vendor-provided shim both in phantom and in vivo measurements. Within the examined multicompartment phantom, the B0 variations could be reduced by up to 77\textpercent using the 23Na MRI\textendashbased shim. In human head, B0 variations were reduced by approximately 50\textpercent using an acquisition time of 15 seconds for the 23Na B0 maps and only 1 iteration of B0 shimming. Conclusion The 23Na MRI\textendashbased localized B0 shimming is possible at 7 T within clinically acceptable acquisition durations (\textless 1 minute). It was shown that using the proposed 23Na MRI\textendashbased shimming approach, the 23Na image quality at ultrahigh field strength can be strongly improved.}

{The locus coeruleus (LC), or \textquoteleftblue spot\textquoteright, is a small nucleus located deep in the brainstem that provides the far-reaching noradrenergic neurotransmitter system of the brain. This phylogenetically conserved nucleus has proved relatively intractable to full characterization, despite more than 60 years of concerted efforts by investigators. Recently, an array of powerful new neuroscience tools have provided unprecedented access to this elusive nucleus, revealing new levels of organization and function. We are currently at the threshold of major discoveries regarding how this tiny brainstem structure exerts such varied and significant influences over brain function and behaviour. All LC neurons receive inputs related to autonomic arousal, but distinct subpopulations of those neurons can encode specific cognitive processes, presumably through more specific inputs from the forebrain areas. This ability, combined with specific patterns of innervation of target areas and heterogeneity in receptor distributions, suggests that activation of the LC has more specific influences on target networks than had initially been imagined.}

{In a prior publication, we described a previously unknown eye movement phenomenon during the execution of actively performed multiaxial rotations in high level gymnasts. This phenomenon was consistently observed during the phase of fast free flight rotations and was marked by a prolonged and complete suppression of nystagmus and gaze stabilizing "environment referenced eye movements" (EREM; such as the vestibulo-ocular reflex, optokinetic reflex, smooth pursuit and others). Instead, these eye movements were coupled with intersegmental body movements. We have therefore called it "spinal motor-coupled eye movements" (SCEM) and have interpreted the phenomenon to likely be caused by anti-compensatory functions of more proprioceptive mediated reflexes and perhaps other mechanisms (e.g., top-down regulation as part of a motor plan) to effectively cope with a new-orientation in space, undisturbed by EREM functions. In the phase before landing, the phenomenon was replaced again by the known gaze-stabilizing EREM functions. The present study specifically evaluated long-term measures of vestibulo-ocular reflex functions (VOR) in high level gymnasts and controls during both passively driven monoaxial rotations and context-specific multiaxial somersault simulations in a vestibular lab. This approach provided further insights into the possible roles of adaptive or mental influences concerning the VOR function and how they are associated with the described phenomenon of SCEM. Results showed high inter-individual variability of VOR function in both gymnasts and controls, but no systematic adaptation of the VOR in gymnasts, neither compared to controls nor over a period of three years. This might generally support the hypothesis that the phenomenon of SCEM might indeed be driven more by proprioceptively mediated and situationally dominant eye movement functions than by adaptative processes of the VOR.}

{Purpose: A supervised learning framework is proposed to automatically generate MR sequences and corresponding reconstruction without human knowledge on MR strategies. This enables a target-based optimization from scratch, as well as exploration of novel and flexible MR sequence strategies. Methods: The entire scanning and reconstruction process is simulated end-to-end in terms of RF events, gradient moment events in x and y, and delay times, acting on the input model spin system given in terms of proton density, T1 and T2, and B0. As proof of concept we use both conventional MR images but also binary masks and T1 maps as a target and optimize from scratch using the loss defined by data fidelity, SAR, and scan time. Results: In a first attempt, MRzero learns all gradient and RF events from zero, and is able to generate the aimed at target image. Appending a neural network layer to the reconstruction module also arbitrary targets were learned successfully. Experiments could be translated to image acquisition at a real system (3T Siemens, PRISMA) and could be verified in measurements of phantoms and the human brain in vivo. Discussion/Conclusion: We have developed a fully automated MR sequence generator based on Bloch equation simulations and supervised learning. While we focus on the method herein, having such a differentiable digital MR twin at hand paves the way to a novel way of generating MR sequence and reconstruction solely governed by the target provided, which can be a certain MR contrast, but the possibilities for targets are limitless, e.g. quantification, segmentation, as well as contrasts of other image modalities.}

{Bridged polymacrocyclic ligands featured by structurally different cages offer the possibility of coordinating multiple trivalent lanthanide ions, giving rise to the exploitation of their different physicochemical properties, e.g., multimodal detection for molecular imaging purposes. Intrigued by the complementary properties of optical and MR-based image capturing modalities, we report the synthesis and characterization of the polymetallic Ln(III)-based chelate comprised of two DOTA-amide-based ligands (DOTA-1,4,7,10-tetraazacyclododecane-1,4,7,10-tetraacetic acid) bridged via 1,10-diaza-18-crown-6 (DA18C6) motif. The DOTA-amide moieties and the DA18C6 were used to chelate two Eu(III) ions and one Tb(III) ion, respectively, resulting in a multinuclear heterometallic complex Eu2LTb. The bimetallic complex without Tb(III), Eu2L, displayed a strong paramagnetic chemical exchange saturation transfer (paraCEST) effect. Notably, the luminescence spectra of Eu2LTb featured mixed emission including the characteristic bands of Eu(III) and Tb(III). The advantageous features of the complex Eu2LTb opens new possibilities for the future design of bimodal probes and their potential applicability in CEST MR and optical imaging.}

{The Max Planck Sustainability Network (MPSN) is a grassroots network within the Max Planck Society (MPG), with \textasciitilde370 members from over 60 Max Planck research institutes, aiming to support sustainability within a German science association committed to fundamental research. The MPSN has adopted the twin goals of making research practices within the overall MPG more sustainable and of supporting local Sustainability Groups in making research practices at their individual institutes more sustainable. The MPSN counts members from diverse backgrounds, regarding both academic field of expertise and roles within the MPG. Its activities focus around Energy, Mobility, Supplies and Waste, Biodiversity and Food, with the ambition to assess impact and expense of each proposed measure. The network\textquotesingles long-term vision is to make research more sustainable and to serve as a role model that inspires other scientific organizations to become sustainable and to optimize the operation of research and administration, which require both individual and structural changes.}

The optogenetically driven manipulation of circuit-specific activity has been very successful to enable functional causality studies in animals, but its global effect on the brain is rarely reported. Optical fiber-mediated optogenetic activation and neuronal Ca2+ recording in combination with fMRI provide a multi-modal fMRI platform with cross-scale brain dynamic mapping schemes, which can elucidate network activity upon circuit-specific optogenetic activation. However, despite highly promising prospects in animal brain research, there are still methodological and conceptual deficiencies, e.g., off-target effects and antidromic activity effects, which remain challenging for the current state of the art. To overcome these difficulties, this thesis describes two technical advances applied at the multi-modal fMRI platform, bridging the methodological and conceptual gap in optogenetics, brain function and animal behavior. First, an MRI-guided robotic arm (MgRA) is developed to increase the target accuracy for optogenetic manipulation or microinjection at the multi-modal fMRI platform, merging fMRI with concurrent deep brain optogenetics in rats. The 4-degrees-of-freedom MgRA allows high precision (50 $\mu$m per step) and sufficient mobility range (10 mm in the ventral-dorsal, rostral-caudal and medial-lateral directions) to manipulate fiber optic or injection needles into the brain in real time and provide high flexibility for multi-site targeting along the trajectory, which shows a clear advantage over the standard stereotaxic-based implantation strategy. Second, the multi-modal fMRI platform provides a specific calcium amplitude-based correlation analysis to study corpus callosum (CC)-mediated brain-wide network dynamics with taking antidromic activity effect into consideration. Since the callosal fibers are reciprocally projecting to two hemispheres, bilateral ortho-vs. antidromically evoked neural activity is difficult to disentangle. Here we not only detected strong antidromic activity, but also detailed temporal dynamics through CC-mediated orthodromic inhibitory activity. The calcium amplitude-based correlation map was created to reveal the brain-wide inhibitory effects from the CC-specific optogenetic stimulation. Last, this multi-modal fMRI platform was used to acquire the optogenetically driven neuronal Ca2+ with single-vessel BOLD and cerebral-blood-volume weighted signal from individual venules and arterioles, respectively, in the hippocampus. We characterized distinct spatiotemporal patterns of hippocampal hemodynamic responses that were correlated to the optogenetically evoked Ca2+ events and further demonstrated the significantly reduced neurovascular coupling (NVC) efficiency upon spreading depression-like Ca2+ events. These results provide a direct measure of the NVC function at varied hippocampal states in animal models. Overall, the technical advances described in this thesis demonstrate the powerful multi-modal fMRI platform to map, analyze and characterize the dynamic brain function across multiple scales and underscore the caution to interpret circuit-specific regulatory mechanisms underlying behavioral or functional outcomes with optogenetic tools.

{Humans always wanted to go faster and higher than their own legs could carry them. This led them to invent numerous types of vehicles to move fast over land, water and air. As training how to handle such vehicles and testing new developments can be dangerous and costly, vehicle motion simulators were invented. Motion-based simulators in particular, combine visual and physical motion cues to provide occupants with a feeling of being in the real vehicle. While visual cues are generally not limited in amplitude, physical cues certainly are, due to the limited simulator motion space. A motion cueing algorithm (MCA) is used to map the vehicle motions onto the simulator motion space. This mapping inherently creates mismatches between the visual and physical motion cues. Due to imperfections in the human perceptual system, not all visual/physical cueing mismatches are perceived. However, if a mismatch is perceived, it can impair the simulation realism and even cause simulator sickness. For MCA design, a good understanding of when mismatches are perceived, and ways to prevent these from occurring, are therefore essential. In this thesis a data-driven approach, using continuous subjective measures of the time-varying Perceived Motion Incongruence (PMI), is adopted. PMI in this case refers to the effect that perceived mismatches between visual and physical motion cues have on the resulting simulator realism. The main goal of this thesis was to develop an MCA-independent off-line prediction method for time-varying PMI during vehicle motion simulation, with the aim of improving motion cueing quality. To this end, a complete roadmap, describing how to measure and model PMI and how to apply such models to predict and minimize PMI in motion simulations is presented. Results from several human-in-the-loop experiments are used to demonstrate the potential of this novel approach.}

{Humans always wanted to go faster and higher than their own legs could carry them, leading them to invent numerous types of vehicles to move fast over land, water and air. As training how to handle such vehicles and testing new developments can be dangerous and costly, vehicle motion simulators were invented. Motion-based simulators in particular, combine visual and physical motion cues to provide occupants with a feeling of being in the real vehicle. While visual cues are generally not limited in amplitude, physical cues certainly are, due to the limited simulator motion space. A motion cueing algorithm (MCA) is used to map the vehicle motions onto the simulator motion space. This mapping inherently creates mismatches between the visual and physical motion cues. Due to imperfections in the human perceptual system, not all visual/physical cueing mismatches are perceived. However, if a mismatch is perceived, it can impair the simulation realism and even cause simulator sickness. For MCA design, a good understanding of when mismatches are perceived, and ways to prevent these from occurring, are therefore essential. In this thesis a data-driven approach, using continuous subjective measures of the time-varying Perceived Motion Incongruence (PMI), is adopted. PMI in this case refers to the effect that perceived mismatches between visual and physical motion cues have on the resulting simulator realism. The main goal of this thesis was to develop an MCA-independent off-line prediction method for time-varying PMI during vehicle motion simulation, with the aim of improving motion cueing quality. To this end, a complete roadmap, describing how to measure and model PMI and how to apply such models to predict and minimize PMI in motion simulations is presented. Results from several human-in-the-loop experiments are used to demonstrate the potential of this novel approach.}

{Animal behavior provides context for understanding disease models and physiology. However, that behavior is often characterized subjectively, creating opportunity for misinterpretation and misunderstanding. For example, spatial alternation tasks are treated as paradigmatic tools for examining memory; however, that link is actually an assumption. To test this assumption, we simulated a reinforcement learning agent equipped with a perfect memory process. We found that it learns a simple spatial alternation task more slowly and makes different errors than a group of male rats, illustrating that memory alone may not be sufficient to capture the behavior. We demonstrate that incorporating spatial biases permits rapid learning and enables the model to fit rodent behavior accurately. Our results suggest that even simple spatial alternation behaviors reflect multiple cognitive processes that need to be taken into account when studying animal behavior.Significance statementMemory is a critical function for cognition whose impairment has significant clinical consequences. Experimental systems aimed at testing various sorts of memory are therefore also central. However, experimental designs to test memory are typically based on intuition about the underlying processes. We tested this using a popular system: a spatial alternation task. Using behavioral modeling, we show that the most straightforward intuition that these tasks just probe spatial memory, does not account for the speed with which rats learn or the types of errors they make. Only when memory-independent dynamic spatial preferences are added can the model learn like the rats. This highlights the importance of respecting the complexity of animal behavior to interpret neural function and validate disease models.}

{Objective: To determine whether cervical cord levels of metabolites are associated with pain sensation after spinal cord injury (SCI), we performed magnetic resonance spectroscopy in SCI patients with and without neuropathic pain (NP). Methods: Cervical cord single-voxel spectroscopic data of 24 SCI patients (14 with NP, 10 pain-free) and 21 healthy controls were acquired at C2/3 to investigate metabolite ratios associated with neuroinflammation (choline-containing compounds to myo-inositol (tCho/mI)) and neurodegeneration (total N-acetylaspartate to myo-inositol (tNAA/mI)). NP levels were measured and Spearman\textquoterights correlation tests assessed associations between metabolite levels, cord atrophy, and pin-prick score. Results: In patients with NP, tCho/mI levels were increased (p\textequals0.024) compared to pain-free patients and negatively related to cord atrophy (p\textequals0.006, r\textequals0.714). Better pin-prick score was associated with higher tCho/mI levels (p\textequals0.032, r\textequals0.574). In pain-free patients, tCho/mI levels were not related to cord atrophy (p\textequals0.881, r\textequals0.055) or pin-prick score (p\textequals0.676, r\textequals0.152). tNAA/mI levels were similar in both patient groups (p\textequals0.396) and were not associated with pin-prick score in patients with NP (p\textequals0.405, r\textequals0.242) and pain-free patients (p\textequals0.117, r\textequals0.527). Conclusions: Neuroinflammatory metabolite levels (i.e. tCho/mI) were elevated in patients with NP; its magnitude being associated with less cord atrophy and greater pain sensation (e.g. pin-prick score). This suggests that patients with NP have more residual spinal tissue and greater metabolite turnover than pain-free patients. Neurodegenerative metabolite levels (i.e. tNAA/mI) were associated with greater cord atrophy, but unrelated to NP. Identifying the metabolic NP signature provides new NP treatment targets and could improve patient stratification in interventional trials. Classification of Evidence: This study provides Class II evidence that levels of MR-spectroscopy-identified metabolites of neuroinflammation were elevated in SCI patients with NP compared to those without NP.}

{A method of processing magnetic resonance (MR) data of a sample under investigation, includes the steps of providing the MR data being collected with an MRI scanner apparatus, and subjecting the MR data to a multi-parameter nonlinear regression procedure being based on a non-linear MR model and employing a set of input parameters, wherein the regression procedure results in creating a parameter map of model parameters of the sample, wherein the input parameters (initial values and possibly boundaries) of the regression procedure are estimated by a machine learning based estimation procedure applied to the MR data. The machine learning based estimation procedure preferably includes at least one of at least one neural network and a support vector machine. Furthermore, an MRI scanner apparatus is described.}

{Purpose Standard relaxation time quantification using phase-cycled balanced steady-state free precession (bSSFP), eg, motion-insensitive rapid configuration relaxometry (MIRACLE), is subject to a considerable underestimation of tissue T1 and T2 due to asymmetric intra-voxel frequency distributions. In this work, an artificial neural network (ANN) fitting approach is proposed to simultaneously extract accurate reference relaxation times (T1, T2) and robust field map estimates ( urn:x-wiley:07403194:media:mrm28325:mrm28325-math-0001 , $\Delta$B0) from the bSSFP profile. Methods Whole-brain bSSFP data acquired at 3T were used for the training of a feedforward ANN with N \textequals 12, 6, and 4 phase-cycles. The magnitude and phase of the Fourier transformed complex bSSFP frequency response served as input and the multi-parametric reference set [T1, T2, urn:x-wiley:07403194:media:mrm28325:mrm28325-math-0002 , $\Delta$B0] as target. The ANN predicted relaxation times were validated against the target and MIRACLE. Results The ANN prediction of T1 and T2 for trained and untrained data agreed well with the reference, even for only four acquired phase-cycles. In contrast, relaxometry based on 4-point MIRACLE was prone to severe off-resonance-related artifacts. ANN predicted urn:x-wiley:07403194:media:mrm28325:mrm28325-math-0003 and $\Delta$B0 maps showed the expected spatial inhomogeneity patterns in high agreement with the reference measurements for 12-point, 6-point, and 4-point bSSFP phase-cycling schemes. Conclusion ANNs show promise to provide accurate brain tissue T1 and T2 values as well as reliable field map estimates. Moreover, the bSSFP acquisition can be accelerated by reducing the number of phase-cycles while still delivering robust T1, T2, urn:x-wiley:07403194:media:mrm28325:mrm28325-math-0004 , and $\Delta$B0 estimates.}

{Background: Deficits in emotion recognition have been repeatedly documented in patients diagnosed with attention-deficit/hyperactivity disorder (ADHD), but their neural basis is unknown so far. Methods: In the current study, adult patients with ADHD (n \textequals 44) and healthy control subjects (n \textequals 43) underwent functional magnetic resonance imaging during explicit emotion recognition of stimuli expressing affective information in face, voice, or face-voice combinations. The employed experimental paradigm allowed us to delineate areas for processing audiovisual information based on their functional activation profile, including the bilateral posterior superior temporal gyrus/middle temporal gyrus, amygdala, medial prefrontal cortex, and precuneus, as well as the right posterior thalamus. Results: As expected, unbiased hit rates for correct classification of the expressed emotions were lower in patients with ADHD than in healthy control subjects irrespective of the presented sensory modality. This deficit at a behavioral level was accompanied by lower activation in patients with ADHD versus healthy control subjects in the cortex adjacent to the right superior temporal gyrus/middle temporal gyrus and the right posterior thalamus, which represent key areas for processing socially relevant signals and their integration across modalities. A cortical region adjacent to the right posterior superior temporal gyrus was the only brain region that showed a significant correlation between brain activation and emotion identification performance. Conclusions: Altogether, these results provide the first evidence for a potential neural substrate of the observed impairments in emotion recognition in adults with ADHD.}

{In a random-dot stereogram (RDS), depth percepts of object surfaces are generated using left-eye and right-eye images that comprise interocularly corresponding random black and white dots. The spatial disparities between the corresponding dots determine the surface depths. If the dots are anti-correlated, such that a black dot in one monocular image corresponds to a white dot in the other, disparity tuned neurons in the primary visual cortex (V1) respond as if their preferred disparities become non-preferred and vice versa, reversing the disparity signs reported to higher visual areas. Humans can perceive this illusory reversed depth in peripheral but not central visual field (Zhaoping \& Ackerman 2018), confirming a prediction (Zhaoping 2017) that feedback from higher visual areas to V1, for analysis-by-synthesis in recognition to veto the reversed depth signals from V1 for violating internal knowledges about the visual world, is weaker peripherally. The current study obtained fMRI responses to the RDSs across the visual hierarchy. A linear decoder is trained to recognize the depth order of a disk against background in correlated RDSs using fMRI activity patterns of a brain region in response to such RDSs. If the decoding performance is better than chance after training, we apply the decoder to the fMRI activity patterns in response to the anti-correlated RDSs to see whether it better reports the reversed than non-reversed depth, and if so, then the brain region is said to signal reversed depth. Reversed depth signals were more likely found in higher (e.g., parietal) than lower (e.g., V1, V2) visual areas, more likely for peripherally viewed RDSs, and more likely for observers who perceived reversed depth (peripherally). Some brain areas, e.g., hV4 and LO, contain the reversed depth signals in central view even though observers could not perceive them, particularly among observers who can perceive reversed depth peripherally.}

{Prosocial behaviors are hypothesized to require socio-cognitive and empathic abilities-engaging brain regions attributed to the mentalizing and empathy brain networks. Here, we tested this hypothesis with a coordinate-based meta-analysis of 600 neuroimaging studies on prosociality, mentalizing and empathy ($\sim$12,000 individuals). We showed that brain areas recruited by prosocial behaviors only partially overlap with the mentalizing (dorsal posterior cingulate cortex) and empathy networks (middle cingulate cortex). Additionally, the dorsolateral and ventromedial prefrontal cortices were preferentially activated by prosocial behaviors. Analyses on the functional connectivity profile and functional roles of the neural patterns underlying prosociality revealed that in addition to socio-cognitive and empathic processes, prosocial behaviors further involve evaluation processes and action planning, likely to select the action sequence that best satisfies another person\textquotesingles needs. By characterizing the multidimensional construct of prosociality at the neural level, we provide insights that may support a better understanding of normal and abnormal social cognition (e.g., psychopathy).}

Neural circuits in the brain have distinct and highly conserved ratios of excitatory and inhibitory neurons. There is typically about 20-30{\textpercent} of inhibitory neurons in the cortex and hippocampus. The percentage remains unchanged throughout the lifespan of an animal. The role of such a specific proportion in network dynamics remains unclear. To investigate this question, we designed a novel experimental platform that allowed us to reliably isolate inhibitory neurons from mice hippocampus and culture networks with different excitatory/inhibitory ratios. Cultures with a broad range of E/I ratios maintained stable spontaneous bursting dynamics, which we further characterized by looking at inter-burst intervals. Cultures with 10-80{\textpercent} of inhibitory neurons showed similar mean inter-burst intervals, whereas cultures with extreme 0{\textpercent} and 100{\textpercent} of inhibitory neurons developed longer inter-burst intervals. The coefficient of variation of inter-burst intervals grew with the number of inhibitory neurons. To link the network properties and bursting dynamics, we fit a network of leaky integrate-and-fire neurons with spike-frequency adaptation and different ratios of excitatory and inhibitory neurons to the experimental data. The results demonstrate that a wide range of parameters may lead to the bursting dynamics observed in vitro. However, the number of inhibitory connections in fitted networks typically stayed proportional to the number of excitatory connections. This suggests the hypothesis that networks adapt to an unusual number of inhibitory cells by balancing E/I connectivity. We further validated this hypothesis by single-cell measurements in patch-clamp experiments. After fitting the model to the recorded activity we pharmacologically blocked inhibitory receptors in different cultures and correspondingly reduced the inhibitory strength in the model. Without refitting, the increase of the mean inter-burst intervals observed in vitro was precisely matched by changes in silico bursting. We demonstrate analytically that network bursting originates from the interaction of fast inhibition and slow spike-frequency adaptation in the context of noise-driven bistable dynamics. Thus, combining measurements of collective and single-cell activity with network modeling, we identify the main mechanisms underlying the spontaneous activity of cultured hippocampal networks and show that they adapt to different numbers of inhibitory neurons by keeping the E/I connections balanced.

{Sex differences in behavioural patterns of drug abuse and dependence have been hypothesized to be a consequence of sexual dimorphisms in brain pathways, particularly within the dopaminergic reward circuitry. Yet, how potential sex differences are manifested at a neurochemical level remains unclear. Here, we use a meta-analysis approach to investigate whether animal studies robustly indicate a different regulation of striatal dopamine transmission in males and females. Data from 39 microdialysis experiments on female rats (n \textequals 676) were extracted and statistically compared with data from 1,523 male rats. All drugs of abuse, independent of their molecular mechanisms of action, notably increase extracellular dopamine concentrations in the nucleus accumbens (NAc) and caudate putamen (CPu). No significant sex differences in basal levels or in dopaminergic response to drugs of abuse were found. However, basal dopamine levels in CPu (but not NAc) were significantly altered by ovariectomy. In conclusion, there are no sex-dependent differences in basal dopamine levels within the NAc and CPu. Previously reported sex differences in the CPu seem to be a result of ovariectomy and may only to a lesser, non-significant degree be attributed to a sexual duality.}

{Identifying the neurons that control global brain state has been a fundamental topic of research that has largely focused on diffusely-projecting neuromodulatory centers, such as the locus coeruleus (LC). This noradrenergic brain stem nucleus, which projects throughout the forebrain, is thought to act as an \textquotedblleftundifferentiated state controller\textquotedblright across all forebrain targets because LC neurons spike synchronously. However, recent work demonstrated ensembles in the LC and therefore made targeted neuromodulation a possibility. In order to demonstrate that LC ensembles cause targeted neuromodulation, it is necessary to resolve LC ensemble dynamics over time in relation to ongoing cortical states. Here, we used non-negative matrix factorization on LC single unit recordings to investigate the spatial and temporal properties of ensemble activation patterns. We assessed the potential for targeted neuromodulation of the prefrontal cortex (PFC) using LC ensemble activity-triggered local field potential (LFP) power spectrograms. We analyzed 285 single units recorded from 15 urethane-anesthetized rats (range of 5 to 34 simultaneously recorded units). LC ensembles became active at different times. Analysis of auto-correlograms and ensemble-pair cross-correlograms demonstrated that self- and lateral-inhibition of activity is a property of LC ensembles, which may contribute to their sparse activity. Neuromodulatory effects on cortical state were diverse across ensembles. We observed four types of ensemble-triggered LFP spectrograms in the PFC. These results demonstrate that the LC is capable of differentiated neuromodulation of its forebrain targets by dynamic firing patterns across subsets of LC neurons.}

{The Locus Coeruleus (LC), a noradrenergic brain stem nucleus projecting throughout the forebrain, is thought to act as an undifferentiated state controller across all forebrain targets because LC neurons spike synchronously [1, 2]. However, recent work demonstrated ensembles in the LC and therefore made targeted neuromodulation a possibility [3]. This recent study used graph theory to reveal a static snapshot of LC ensembles. In order to now demonstrate that LC ensembles cause targeted neuromodulation, it is necessary to resolve LC ensemble dynamics over time in relation to ongoing cortical states.}

{Learning meaningful and compact representations with structurally disentangled semantic aspects is considered to be of key importance in representation learning. Since real-world data is notoriously costly to collect, many recent state-of-the-art disentanglement models have heavily relied on synthetic toy data-sets. In this paper, we propose a novel data-set which consists of over 1 million images of physical 3D objects with seven factors of variation, such as object color, shape, size and position. In order to be able to control all the factors of variation precisely, we built an experimental platform where the objects are being moved by a robotic arm. In addition, we provide two more datasets which consist of simulations of the experimental setup. These datasets provide for the first time the possibility to systematically investigate how well different disentanglement methods perform on real data in comparison to simulation, and how simulated data can be leveraged to build better representations of the real world. We provide a first experimental study of these questions and our results indicate that learned models transfer poorly, but that model and hyperparameter selection is an effective means of transferring information to the real world.}

{Magnetization transfer (MT), reflecting the exchange of magnetization between mobile and bound protons, has shown good potential for the diagnosis and prognosis of various neurological disorders, such as multiple sclerosis. Frequently, MT effects are assessed by measuring the contrast between two scans performed with and without saturation of the bound pool protons. Evidently, saturation is affected by B1 inhomogeneity and should be accounted for. In this work, we report on a very rapid one-minute whole-brain magnetization transfer ratio (MTR) imaging method offering intrinsic B1-correction.}

{Accurate reproduction of the feeling of motion is the primary goal of motion simulation. The subjective experience of self-motion results from the processing of visual and inertial sensory information in the human central nervous system (CNS). In contrast to the real experience, the simulation capabilities are restricted by the motion simulator workspace, which makes the exact reproduction of the inertial sensory input generally impossible. It is possible,however, to minimize the mismatch between the desired and the actual inertial signal, while satisfying the constraints imposed by the motion simulator. This naturally leads to the formulation of the motion simulation problem as a constrained optimal control problem (OCP).If this OCP is solved in real-time on a moving horizon, we enter the field of model predictive control (MPC). The MPC-based motion simulation is computationally expensive due to the (large) size of the problem and the need to take the generally non-linear dynamics of the simulator into account. The solution time must be small enough to achieve the desired control rate. Use of appropriate algorithms and a thoughtful approach to software development are needed to make the MPC implementation real-time capable.Another important problem is the offline design of motion simulators. The optimal design parameters can be obtained from an optimization problem, which can be solved simultaneously with finding optimal control inputs in an offline context.This work focuses on developing efficient approaches for solving optimization problems arising in the field of motion simulation. In the offline context, a method to simultaneously optimize simulator trajectories and design parameters is developed. In the online context, real-time capable implementations for two motion simulators at the Max Planck Institute for Biological Cybernetics in T\"ubingen, Germany are created and their performance is analyzed. One of the implementations is experimentally validated by measuring the inertial signal tracking error for realistic motion simulation scenarios.The efficiency of the software implementation plays a critical role in real-time MPC. In order to simplify the development of MPC controllers and to be able to reuse algorithms, a number of software packages exist. Such software packages must be efficient, convenient to use, and provide a sufficient degree of control over the implementation details to the user. To satisfy these conflicting demands, the tmpc library was developed in the context of this work and was used to implement the real-time controllers for the two motion simulators. The library is released as open source in the hope that it will be useful for developing further real-time MPC applications.}

{n in vivo MRSI spatial inhomogeneities of the main magnetic field cause spectral line broadening and location dependent frequency shifts in the spectra. An over-discretized reconstruction method is applied to existing prostate 1H-MRSI data, consisting of spatial over-discretization of the MRSI grid, frequency shift correction according to a high resolution B0map and weighted spatial averaging to the target resolution. Intravoxel B0 variations are corrected for and resampling to a new target resolution results in improved localization of signals. This could lead to detection of signals with lower spectral intensity and to improved visualization of smaller cancer lesions in the prostate.}

{Many studies have demonstrated that we can identify a familiar face on an image much better than an unfamiliar one, especially when various degradations or changes (e.g., image distortions or blurring, new illuminations) have been applied, but few have asked how different types of facial information from familiar faces are stored in memory. Here we investigated how well we remember personally familiar faces in terms of their identity, gender, and race. In 3 experiments, based on the faces personally familiar to our participants, we created sets of face morphs that parametrically varied the faces in terms of identity, sex, or race using a 3-dimensional morphable face model. For each familiar face, we presented those face morphs together with the original face and asked participants to pick the correct "real" face among morph distracters in each set. They were instructed to pick the face that most closely resembled their memory of that familiar person. We found that participants excelled in retrieving the correct familiar faces among the distracters when the faces were manipulated in terms of their idiosyncratic features (their identity information), but they were less sensitive to changes that occurred along the gender and race continuum. Image similarity analyses indicate that the observed difference cannot be attributed to different levels of image similarity between manipulations. These findings demonstrate that idiosyncratic and categorical face information is represented differently in memory, even for the faces of people we are very familiar with. Implications to current models of face recognition are discussed.}