ArXiv最新文献

Previous studies have compared neural activities in the visual cortex to representations in deep neural networks trained on image classification. Interestingly, while some suggest that their representations are highly similar, others argued the opposite. Here, we propose a new approach to characterize the similarity of the decision strategies of two observers (models or brains) using decision variable correlation (DVC). DVC quantifies the image-by-image correlation between the decoded decisions based on the internal neural representations in a classification task. Thus, it can capture task-relevant information rather than general representational alignment. We evaluate DVC using monkey V4/IT recordings and network models trained on image classification tasks. We find that model-model similarity is comparable to monkey-monkey similarity, whereas model-monkey similarity is consistently lower. Strikingly, DVC decreases with increasing network performance on ImageNet-1k. Adversarial training does not improve model-monkey similarity in task-relevant dimensions assessed using DVC, although it markedly increases the model-model similarity. Similarly, pre-training on larger datasets does not improve model-monkey similarity. These results suggest a divergence between the task-relevant representations in monkey V4/IT and those learned by models trained on image classification tasks.

Common deep learning approaches for antibody engineering focus on modeling the marginal distribution of sequences. By treating sequences as independent samples, however, these methods overlook affinity maturation as a rich and largely untapped source of information about the evolutionary process by which antibodies explore the underlying fitness landscape. In contrast, classical phylogenetic models explicitly represent evolutionary dynamics but lack the expressivity to capture complex epistatic interactions. We bridge this gap with CoSiNE, a continuous-time Markov chain parameterized by a deep neural network. Mathematically, we prove that CoSiNE provides a first-order approximation to the intractable sequential point mutation process, capturing epistatic effects with an error bound that is quadratic in branch length. Empirically, CoSiNE outperforms state-of-the-art language models in zero-shot variant effect prediction by explicitly disentangling selection from context-dependent somatic hypermutation. Finally, we introduce Guided Gillespie, a classifier-guided sampling scheme that steers CoSiNE at inference time, enabling efficient optimization of antibody binding affinity toward specific antigens.

In the last century, most sensorimotor studies of cortical neurons relied on average firing rates. Rate coding is efficient for fast sensorimotor processing that occurs within a few seconds. Much less is known about the neural mechanisms underlying long-term working memory with a time scale of hours (Ericsson and Kintsch, 1995). Cognitive states may not have sensory or motor correlates. For example, you can sit in a quiet room making plans without moving or sensory processing. You can also make plans while out walking. This suggests that the neural substrate for cognitive states neither depends on nor interferes with ongoing sensorimotor brain activity. In this perspective, I make the case for a possible second tier of neural activity that coexists with the well-established sensorimotor tier, based on coordinated spike-timing activity. The discovery of millisecond-precision spike initiation in cortical neurons was unexpected (Mainen and Sejnowski, 1995). Even more striking was the precision of spiking in vivo, in response to rapidly fluctuating sensory inputs, suggesting that neural circuits could preserve and manipulate sensory information through spike timing. High temporal resolution can also mediate spike-timing-dependent plasticity (STDP) by controlling the relative timing of presynaptic and postsynaptic spikes at the millisecond scale. Cortical traveling waves with high temporal precision are observed across many frequency bands. They can plausibly trigger STDP that lasts for hours in cortical neurons. This temporary cortical network, riding astride the long-term sensorimotor network, could support cognitive processing and long-term working memory.

Neuronal electrical activity underlies human cognition including perception, attention, memory, language, and decision-making. Yet its direct, noninvasive measurement in the living human brain remains a fundamental challenge. Existing neuroimaging techniques, including electroencephalography (EEG), magnetoencephalography (MEG), and functional magnetic resonance imaging (fMRI), are limited by trade-offs in sensitivity and spatial or temporal resolution. Here we propose quantum sensing MRI (qsMRI), a noninvasive approach that enables direct detection of neuronal firing-induced magnetic fields using a clinical MRI system. qsMRI exploits endogenous proton (1H) nuclear spins in water molecules as intrinsic quantum sensors and decodes time-resolved phase information from the free induction decay signals to infer neuronal magnetic fields. We validate qsMRI through simulations, phantom experiments, and human studies at rest and during motor tasks, and provide open experimental procedures to facilitate independent rigorous validation. We further present a case study demonstrating potential applications to neurological disorders. qsMRI represents, to our knowledge, the first-in-human application of quantum sensing on a clinical MRI platform and may lay the foundation for a non-BOLD functional imaging modality capable of probing neuronal firing dynamics in both cortical and deep brain regions.

Background: The claim that Monte Carlo is the most accurate method is a case of misattributed credit. This claim is based on experience with advanced systems MC-NPX, Geant4 and EGS. These systems achieve remarkable performance because they use most accurate physics, not because they use random numbers. The latter simplifies algorithms, but contaminates the solution with random noise. Currently prevalent fast Monte Carlo algorithms retain this worst part while achieving high computing speed at the expense of the best part - accurate physics. We employ an opposite strategy. We develop a Boltzmann solver for protons that retains unchanged the physics of most advanced Monte Carlo systems. We eliminate random noise, because our solution method is deterministic. Our method is also applicable to heavier ions, helium and carbon, for example.

Purpose: To develop a fast and accurate deterministic Boltzmann solver for protons. It calculates dose distributions and fluence spectra. The spectra are needed for biological modelling. The main application is treatment planning of proton beam therapy.

Methods: We solve the Boltzmann transport equation using an iterative procedure. Our algorithm accounts for Coulomb scattering and nuclear reactions. It uses the same physical models, as do the most rigorous Monte Carlo systems. Thereby it achieves the same low level of systematic errors. Our solver does not involve random sampling. The solution is not contaminated by statistical noise. This means that the overall uncertainties of our solver are lower than those realistically achievable with Monte Carlo. Furthermore, our solver is orders of magnitude faster. Its another advantage is that it calculates fluence spectra. They are needed for calculation of relative biological effectiveness, especially when advanced radiobiological models are used that may present a challenge for other algorithms.

Results: We have developed a novel Boltzmann equation solver, have written prototype software, and completed its testing for calculations in water. For 40-220 MeV protons we calculated fluence spectra, depth doses, three-dimensional dose distributions for narrow Gaussian beams. The CPU time was 5-11 ms for depth doses and fluence spectra at multiple depths. Gaussian beam calculations took 31-78 ms. All the calculations were run on a single Intel i7 2.9 GHz CPU. Comparison of our solver with Geant4 showed good agreement for all energies and depths. For the 1%/1 mm $\gamma$ -test the pass rate was 0.95-0.99. In this test, 1% was the difference between our and Geant4 doses at the same point. The test included low dose regions down to 1% of the maximum dose.

Conclusions: Results of the study provide a foundation for achieving a high computing speed with uncompromised accuracy in proton treatment planning systems.

Despite Ultra-High Field MRI (UHF-MRI) being increasingly used in large-scale neuroimaging studies, automatic segmentation and parcellation remain challenging due to signal inhomogeneities, varying contrast and resolution, and the lack of tools optimized for UHF-MRI. Traditional software packages such as FastSurferVINN or SynthSeg ⁺ often yield suboptimal results when applied directly to UHF images, which has limited region-based quantitative analyses. Building upon this need, we propose GOUHFI 2.0, a new implementation of GOUHFI that incorporates greater training data variation and introduces added functionalities, including cortical parcellation and volumetry. GOUHFI 2.0 preserves the contrast- and resolution-agnostic properties of the original toolbox while introducing two independently trained segmentation tasks based on the 3D U-Net architecture. The first network segments brain images of any contrast, resolution or field strength into 35 labels, using the domain randomization approach with a dataset composed of 238 subjects of varied resolutions, field strengths and populations. Using the same training dataset, the second network performs the parcellation of the cortex into 62 labels following the Desikan-Killiany-Tourville (DKT) protocol. When evaluated across multiple datasets, GOUHFI 2.0 demonstrated improved segmentation accuracy relative to the original toolbox, particularly in heterogeneous populations, and its ability to generate reliable cortical parcellations. Additionally, the added integrated volumetry pipeline enabled the derivation of results consistent with those obtained using standard volumetry procedures. In summary, GOUHFI 2.0 offers a comprehensive, contrast- and resolution-agnostic solution for brain segmentation and parcellation across field strengths. This positions GOUHFI 2.0 as a versatile tool for researchers working at UHF-MRI, making it the first Deep Learning (DL) toolbox capable of robust cortical parcellation at UHF-MRI.

Head impacts can cause concussion, but the precise biomechanical conditions that produce injury remain uncertain. Rotational acceleration has long been posited as the primary cause and has guided concussion prevention strategies. Using instrumented mouthguards to record head kinematics of diagnosed concussions, we directly tested this hypothesis and found that linear acceleration predicted injury with greater precision than rotational acceleration, while rotational velocity provided additional predictive value. Injury risk functions derived from these measurements indicated substantial predicted concussion risk during typical impacts to an American football helmet. Introducing a liquid-filled helmet pad designed to attenuate linear acceleration reduced predicted risk by up to 52%. These results indicate that effective concussion prevention requires targeting linear acceleration.

Understanding how neural networks transform inputs into outputs is crucial for interpreting and manipulating their behavior. Most existing approaches analyze internal representations by identifying hidden-layer activation patterns correlated with human-interpretable concepts. Here we take a direct approach to examine how hidden neurons act to drive network outputs. We introduce CODEC (Contribution Decomposition), a method that uses sparse autoencoders to decompose network behavior into sparse motifs of hidden-neuron contributions, revealing causal processes that cannot be determined by analyzing activations alone. Applying CODEC to benchmark image-classification networks, we find that contributions grow in sparsity and dimensionality across layers and, unexpectedly, that they progressively decorrelate positive and negative effects on network outputs. We further show that decomposing contributions into sparse modes enables greater control and interpretation of intermediate layers, supporting both causal manipulations of network output and human-interpretable visualizations of distinct image components that combine to drive that output. Finally, by analyzing state-of-the-art models of neural activity in the vertebrate retina, we demonstrate that CODEC uncovers combinatorial actions of model interneurons and identifies the sources of dynamic receptive fields. Overall, CODEC provides a rich and interpretable framework for understanding how nonlinear computations evolve across hierarchical layers, establishing contribution modes as an informative unit of analysis for mechanistic insights into artificial neural networks.

Estimating brain microstructure has important applications in medicine and neuroscience. Diffusion-weighted magnetic resonance imaging enables measuring microstructure in vivo. Conventional biophysical model fitting can be accurate but is slow and impractical for time-critical clinical use, where machine learning can offer a potential route to rapid estimation. We address the problem of microstructure estimation under arbitrary acquisition protocols where most existing learning-based methods fail due to protocol assumptions, requiring retraining when the protocol changes. We present a graph neural network that represents input data as a point cloud in the 3D space where diffusion-weighted measurements are made and performs rotation-invariant message passing with permutation-invariant pooling, producing fixed-size embeddings that encode microstructure. The inductive biases of our relatively small model were guided by the underlying physics and symmetries of the problem rather than by generic model architectures. Trained on randomised simulated data, our model demonstrates domain generalisation, accurately estimating microstructure from data with unseen real-world protocols without retraining. This approach represents a step towards a "train once, deploy anywhere" architecture, bringing rapid learning-based microstructure mapping closer to clinical deployment.