Publications

Computational models of reinforcement learning (RL) have significantly contributed to our understanding of human behavior and decision-making. Traditional RL models, however, often adopt a linear approach to updating reward expectations, potentially oversimplifying the nuanced relationship between human behavior and rewards. To address these challenges and explore models of RL, we utilized a method of model discovery using equation discovery algorithms. This method, currently used mainly in physics and biology, attempts to capture data by proposing a differential equation from an array of suggested linear and nonlinear functions. Using this method, we were able to identify a model of RL which we termed the Quadratic Q-Weighted model. The model suggests that reward prediction errors obey nonlinear dynamics and exhibit negativity biases, resulting in an underweighting of reward when expectations are low, and an overweighting of the absence of reward when expectations are high. We tested the generalizability of our model by comparing it to classical models used in nine published studies. Our model surpassed traditional models in predictive accuracy across eight out of these nine published datasets, demonstrating not only its generalizability but also its potential to offer insights into the complexities of human learning. This work showcases the integration of a behavioral task with advanced computational methodologies as a potent strategy for uncovering the intricate patterns of human cognition, marking a significant step forward in the development of computational models that are both interpretable and broadly applicable.

Computational phenotyping has emerged as a powerful tool for characterizing individual variability across a variety of cognitive domains. An individuals computational phenotype is defined as a set of mechanistically interpretable parameters obtained from fitting computational models to behavioural data. However, the interpretation of these parameters hinges critically on their psychometric properties, which are rarely studied. To identify the sources governing the temporal variability of the computational phenotype, we carried out a 12-week longitudinal study using a battery of seven tasks that measure aspects of human learning, memory, perception and decision making. To examine the influence of state effects, each week, participants provided reports tracking their mood, habits and daily activities. We developed a dynamic computational phenotyping framework, which allowed us to tease apart the time-varying effects of practice and internal states such as affective valence and arousal. Our results show that many phenotype dimensions covary with practice and affective factors, indicating that what appears to be unreliability may reflect previously unmeasured structure. These results support a fundamentally dynamic understanding of cognitive variability within an individual.

Uncovering the architecture of white matter axons is fundamental to the study of brain networks. We developed a method for quantifying axonal orientations at a resolution of ~15 micrometers. This method is based on the common Nissl staining technique for postmortem histological slices. Nissl staining reveals the spatial organization of glial cells along axons. Using structure tensor analysis, we leveraged this patterned organization to uncover local axonal orientation. We used Nissl-based structure tensor analysis to extract fine details of axonal architecture and demonstrated its applicability in multiple datasets of humans and nonhuman primates. Nissl-based structure tensor analysis can be used to compare fine-grained features of axonal architecture across species and is widely applicable to existing datasets.

Brain development and aging are complex processes that unfold in multiple brain regions simultaneously. Recently, models of brain age prediction have aroused great interest, as these models can potentially help to understand neurological diseases and elucidate basic neurobiological mechanisms. We test whether quantitative magnetic resonance imaging can contribute to such age prediction models. Using R1, the longitudinal rate of relaxation, we explore lifespan dynamics in cortical gray matter. We compare R1 with cortical thickness, a well-established biomarker of brain development and aging. Using 160 healthy individuals (6-81 years old), we found that R1 and cortical thickness predicted age similarly, but the regions contributing to the prediction differed. Next, we characterized R1 development and aging dynamics. Compared with anterior regions, in posterior regions we found an earlier R1 peak but a steeper postpeak decline. We replicate these findings: Firstly, we tested a subset (N = 10) of the original dataset for whom we had additional scans at a lower resolution; and second, we verified the results on an independent dataset (N = 34). Finally, we compared the age prediction models on a subset of 10 patients with multiple sclerosis. The patients are predicted older than their chronological age using R1 but not with cortical thickness.

The claustrum is a thin sheet of neurons enclosed by white matter and situated between the insula and the putamen. It is highly interconnected with sensory, frontal, and subcortical regions. The deep location of the claustrum, with its fine structure, has limited the degree to which it could be studied in vivo. Particularly in humans, identifying the claustrum using magnetic resonance imaging (MRI) is extremely challenging, even manually. Therefore, automatic segmentation of the claustrum is an invaluable step toward enabling extensive and reproducible research of the anatomy and function of the human claustrum. In this study, we developed an automatic algorithm for segmenting the human dorsal claustrum in vivo using high-resolution MRI. Using this algorithm, we segmented the dorsal claustrum bilaterally in 1068 subjects of the Human Connectome Project Young Adult dataset, a publicly available high-resolution MRI dataset. We found good agreement between the automatic and manual segmentations performed by 2 observers in 10 subjects. We demonstrate the use of the segmentation in analyzing the covariation of the dorsal claustrum with other brain regions, in terms of macro- and microstructure. We identified several covariance networks associated with the dorsal claustrum. We provide an online repository of 1068 bilateral dorsal claustrum segmentations.

The association fibers of the superior longitudinal fasciculus (SLF) connect parietal and frontal cortical regions in the human brain. The SLF comprises of three distinct sub-bundles, each presenting a different anatomical trajectory, and specific functional roles. Nevertheless, in vivo studies of the SLF often consider the entire SLF complex as a single entity. In this work, we suggest a data-driven approach that relies on microstructure measurements for separating SLF-III from the rest of the SLF. We apply the SLF-III separation procedure in three independent datasets using parameters of diffusion MRI (fractional anisotropy), as well as relaxometry-based parameters (T1, T2, T2* and T2-weighted/T1-weighted). We show that the proposed procedure is reproducible across datasets and tractography algorithms. Finally, we suggest that differential crossing with different white-matter tracts is the source of the distinct MRI signatures of SLF-II and SLF-III.

The vertical occipital fasciculus (VOF) is a white-matter tract that connects the ventral and dorsal visual streams. The precise borders of the VOF have been a matter of dispute since its discovery in the 19th century. The presence of an adjacent vertical pathway, the posterior arcuate fasciculus, makes it especially hard to determine the anterior extent of the VOF. By integrating diffusion MRI tractography with quantitative T1 mapping we found that the vertical streamlines originating in the ventral occipito-temporal cortex show a pattern of lower T1 in more posterior streamlines. We used this pattern to develop an automatic procedure for VOF identification based on a sharp increase in the streamline T1 signature along the posterior-anterior axis. We studied the cortical endpoints of the VOF and their relation to known cytoarchitectonic and functional divisions of the cortex. These results show that multi-modal MRI information, which characterizes local tissue microstructure such as myelination, can be used to delineate white-matter tracts in vivo.

The arcuate fasciculi are white-matter pathways that connect frontal and temporal lobes in each hemisphere. The arcuate plays a key role in the language network and is believed to be left-lateralized, in line with left hemisphere dominance for language. Measuring the arcuate in vivo requires diffusion magnetic resonance imagingbased tractography, but asymmetry of the in vivo arcuate is not always reliably detected in previous studies. It is unknown how the choice of tractography algorithm, with each method's freedoms, constraints, and vulnerabilities to false-positive and -negative errors, impacts findings of arcuate asymmetry. Here, we identify the arcuate in two independent datasets using a number of tractography strategies and methodological constraints, and assess their impact on estimates of arcuate laterality. We test three tractography methods: a deterministic, a probabilistic, and a tractography-evaluation (LiFE) algorithm. We extract the arcuate from the whole-brain tractogram, and compare it to an arcuate bundle constrained even further by selecting only those streamlines that connect to anatomically relevant cortical regions. We test arcuate macrostructure laterality, and also evaluate microstructure profiles for properties such as fractional anisotropy and quantitative R1. We find that both tractography choice and implementing the cortical constraints substantially impact estimates of all indices of arcuate laterality. Together, these results emphasize the effect of the tractography pipeline on estimates of arcuate laterality in both macrostructure and microstructure.

Quantitative magnetic resonance imaging (qMRI) is a technique for mapping the physical properties of the underlying tissue using several MR images with different contrasts. To overcome subject motion between the acquired images, it is necessary to register the images to a common reference frame. A drawback of registration is the use of interpolation and resampling techniques, which can introduce artifacts into the interpolated data. These artifacts could have unfavorable effects on the accuracy of the estimated tissue's physical properties. Here, we quantified the error of interpolation and resampling on T1-weighted images and studied its effects on the mapping of the longitudinal relaxation time (T1) using variable flip angles. We simulated T1-weighted images and calculated the transformation error resulting from interpolation and resampling. We found that the error is a function of the image contrast (i.e., flip angle) and of the translation and rotation of the image. Furthermore, we found that the error in the T1-weighted images has a substantial effect on the T1 estimation, of the order of 10% of the signal in the brain's gray and white matter. Hence, minimizing the registration error can enable more accurate in vivo modeling of brain microstructure.

Diffusion MRI tractography is essential for reconstructing white-matter projections in the living human brain. Yet tractography results miss some projections and falsely identify others. A challenging example is the optic radiation (OR) that connects the thalamus and the primary visual cortex. Here, we tested whether OR tractography can be optimized using quantitative T1 mapping. Based on histology, we proposed that myelin-sensitive T1 values along the OR should remain consistently low compared with adjacent white matter. We found that complementary information from the T1 map allows for increasing the specificity of the reconstructed OR tract by eliminating falsely identified projections. This T1-filtering outperforms other, diffusion-based tractography filters. These results provide evidence that the smooth microstructural signature along the tract can be used as constructive input for tractography. Finally, we demonstrate that this approach can be applied in a case of multiple sclerosis, and generalized to the HCP-available MRI measurements. We conclude that multimodal MRI microstructural information can be used to eliminate spurious tractography results in the case of the OR.

In mental time travel (MTT) one is \u201ctraveling\u201d back-and-forth in time, remembering, and imagining events. Despite intensive research regarding memory processes in the hippocampus, it was only recently shown that the hippocampus plays an essential role in encoding the temporal order of events remembered, and therefore plays an important role in MTT. Does it also encode the temporal relations of these events to the remembering self? We asked patients undergoing pre-surgical evaluation with depth electrodes penetrating the temporal lobes bilaterally toward the hippocampus to project themselves in time to a past, future, or present time-point, and then make judgments regarding various events. Classification analysis of intracranial evoked potentials revealed clear temporal dissociation in the left hemisphere between lateral-temporal electrodes, activated at ∼100300ms, and hippocampal electrodes, activated at ∼400600ms. This dissociation may suggest a division of labor in the temporal lobe during self-projection in time, hinting toward the different roles of the lateral-temporal cortex and the hippocampus in MTT and the temporal organization of the related events with respect to the experiencing self.

Religious experiences have long been documented in patients with epilepsy, though their exact underlying neural mechanisms are still unclear. Here, we had the rare opportunity to record a delusional religious conversion in real time in a patient with right temporal lobe epilepsy undergoing continuous video-EEG. In this patient, a messianic revelation experience occurred several hours after a complex partial seizure of temporal origin, compatible with postictal psychosis (PIP). We analyzed the recorded resting-state EEG epochs separately for each of the conventional frequency bands. Topographical analysis of the bandpass filtered EEG epochs revealed increased activity in the low-gamma range (30-40 Hz) during religious conversion compared with activity during the patient's habitual state. The brain generator underlying this activity was localized to the left prefrontal cortex. This suggests that religious conversion in PIP is related to control mechanisms in the prefrontal lobe-related processes rather than medial temporal lobe-related processes.