Abstract Traumatic brain injury (TBI), particularly from explosive blasts, is a major cause of casualties in modern military conflicts. Computational models are an important tool in understanding the underlying biomechanics of TBI but are highly dependent on the mechanical properties of soft tissue to produce accurate results. Reported material properties of brain tissue can vary by several orders of magnitude between studies, and no published set of material parameters exists for porcine brain tissue at strain rates relevant to blast. In this work, brain tissue from the brainstem, cerebellum, and cerebrum of freshly euthanized adolescent male Göttingen minipigs was tested in simple shear and unconfined compression at strain rates ranging from quasi-static (QS) to 300 s−1. Brain tissue showed significant strain rate stiffening in both shear and compression. Minimal differences were seen between different regions of the brain. Both hyperelastic and hyper-viscoelastic constitutive models were fit to experimental stress, considering data from either a single loading mode (unidirectional) or two loading modes together (bidirectional). The unidirectional hyper-viscoelastic models with an Ogden hyperelastic representation and a one-term Prony series best captured the response of brain tissue in all regions and rates. The bidirectional models were generally able to capture the response of the tissue in high-rate shear and all compression modes, but not the QS shear. Our constitutive models describe the first set of material parameters for porcine brain tissue relevant to loading modes and rates seen in blast injury.
more »
« less
This content will become publicly available on October 1, 2024
Bidirectional Mapping with Contrastive Learning on Multimodal Neuroimaging Data
The modeling of the interaction between brain structure and function using deep learning techniques has yielded remarkable success in identifying potential biomarkers for different clinical phenotypes and brain diseases. However, most existing studies focus on one-way mapping, either projecting brain function to brain structure or inversely. This type of unidirectional mapping approach is limited by the fact that it treats the mapping as a one-way task and neglects the intrinsic unity between these two modalities. Moreover, when dealing with the same biological brain, mapping from structure to function and from function to structure yields dissimilar outcomes, highlighting the likelihood of bias in one-way mapping. To address this issue, we propose a novel bidirectional mapping model, named Bidirectional Mapping with Contrastive Learning (BMCL), to reduce the bias between these two unidirectional mappings via ROI-level contrastive learning. We evaluate our framework on clinical phenotype and neurodegenerative disease predictions using two publicly available datasets (HCP and OASIS). Our results demonstrate the superiority of BMCL compared to several state-of-the-art methods.
more »
« less
- NSF-PAR ID:
- 10518866
- Publisher / Repository:
- Lecture Notes in Computer Science
- Date Published:
- Volume:
- 14222
- ISBN:
- 978-3-031-43897-4
- Subject(s) / Keyword(s):
- Bidirectional reconstruction BOLD signals Structural networks Prediction Biomarkers
- Format(s):
- Medium: X
- Location:
- Vancouver, Canada
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Unsupervised learning has recently made exceptional progress because of the development of more effective contrastive learning methods. However, CNNs are prone to depend on low-level features that humans deem non-semantic. This dependency has been conjectured to induce a lack of robustness to image perturbations or domain shift. In this paper, we show that by generating carefully designed negative samples, contrastive learning can learn more robust representations with less dependence on such features. Contrastive learning utilizes positive pairs which preserve semantic information while perturbing superficial features in the training images. Similarly, we propose to generate negative samples in a reversed way, where only the superfluous instead of the semantic features are preserved. We develop two methods, texture-based and patch-based augmentations, to generate negative samples. These samples achieve better generalization, especially under out-of-domain settings. We also analyze our method and the generated texture-based samples, showing that texture features are indispensable in classifying particular ImageNet classes and especially finer classes. We also show that the model bias between texture and shape features favors them differently under different test settings.more » « less
-
In this paper, we study the problem of unsupervised graph representation learning by harnessing the control properties of dynamical networks defined on graphs. Our approach introduces a novel framework for contrastive learning, a widely prevalent technique for unsupervised representation learning. A crucial step in contrastive learning is the creation of ‘augmented’ graphs from the input graphs. Though different from the original graphs, these augmented graphs retain the original graph’s structural characteristics. Here, we propose a unique method for generating these augmented graphs by leveraging the control properties of networks. The core concept revolves around perturbing the original graph to create a new one while preserving the controllability properties specific to networks and graphs. Compared to the existing methods, we demonstrate that this innovative approach enhances the effectiveness of contrastive learning frameworks, leading to superior results regarding the accuracy of the classification tasks. The key innovation lies in our ability to decode the network structure using these control properties, opening new avenues for unsupervised graph representation learning.more » « less
-
more » « less
Abstract The calculation of so‐called “brain age” from structural MRIs has been an emerging biomarker in aging research. Data suggests that discrepancies between chronological age and the predicted age of the brain may be predictive of mortality and morbidity (for review, see Cole, Marioni, Harris, & Deary, 2019). However, with these promising results come technical complexities of how to calculate brain age. Various groups have deployed methods leveraging different statistical approaches, often crafting novel algorithms for assessing this biomarker derived from structural MRIs. There remain many open questions about the reliability, collinearity, and predictive power of different algorithms. Here, we complete a rigorous systematic comparison of three commonly used, previously published brain age algorithms (XGBoost, brainageR, and DeepBrainNet) to serve as a foundation for future applied research. First, using multiple datasets with repeated structural MRI scans, we calculated two metrics of reliability (intraclass correlations and Bland–Altman bias). We then considered correlations between brain age variables, chronological age, biological sex, and image quality. We also calculated the magnitude of collinearity between approaches. Finally, we used machine learning approaches to identify significant predictors across brain age algorithms related to clinical diagnoses of cognitive impairment. Using a large sample (
N = 2557 ), we find all three commonly used brain age algorithms demonstrate excellent reliability (r > .9). We also note that brainageR and DeepBrainNet are reasonably correlated with one another, and that the XGBoost brain age is strongly related to image quality. Finally, and notably, we find that XGBoost brain age calculations were more sensitive to the detection of clinical diagnoses of cognitive impairment. We close this work with recommendations for future research studies focused on brain age. -
Boldyreva, Alexandra ; Kolesnikov, Vladimir (Ed.)Updatable Encryption (UE) and Proxy Re-encryption (PRE) allow re-encrypting a ciphertext from one key to another in the symmetric-key and public-key settings, respectively, without decryption. A longstanding open question has been the following: do unidirectional UE and PRE schemes (where ciphertext re-encryption is permitted in only one direction) necessarily require stronger/more structured assumptions as compared to their bidirectional counterparts? Known constructions of UE and PRE seem to exemplify this “gap” – while bidirectional schemes can be realized as relatively simple extensions of public-key encryption from standard assumptions such as DDH or LWE, unidirectional schemes typically rely on stronger assumptions such as FHE or indistinguishability obfuscation (iO), or highly structured cryptographic tools such as bilinear maps or lattice trapdoors. In this paper, we bridge this gap by showing the first feasibility results for realizing unidirectional UE and PRE from a new generic primitive that we call Key and Plaintext Homomorphic Encryption (KPHE) – a public-key encryption scheme that supports additive homomorphisms on its plaintext and key spaces simultaneously. We show that KPHE can be instantiated from DDH. This yields the first constructions of unidirectional UE and PRE from DDH. Our constructions achieve the strongest notions of post-compromise security in the standard model. Our UE schemes also achieve “backwards-leak directionality” of key updates (a notion we discuss is equivalent, from a security perspective, to that of unidirectionality with no-key updates). Our results establish (somewhat surprisingly) that unidirectional UE and PRE schemes satisfying such strong security notions do not, in fact, require stronger/more structured cryptographic assumptions as compared to bidirectional schemes.more » « less
