Biological validity, test–retest reliability, and behavioral relevance of the single-subject brain volumetric similarity network

NeuroImage · 2026-03-13

The T1-weighted brain magnetic resonance imaging (MRI)-based volumetric similarity network (VSN) offers an advantage in clinical settings due to its ease of acquisition and widespread availability. However, its validity, reliability, and behavioral relevance remain unclear. The present study aimed to assess the reproducibility and utility of the VSN as a foundation for future research and clinical applications. Here, we analyzed three datasets (total N = 354), with two datasets having repeated MR runs (Dataset 1: n = 86; Dataset 2: n = 49) and two having an attention measure (Datasets 1 and 3: n = 219). For each run and participant, the VSN was generated using interregional morphological similarity metrics. We examined whether the VSN reflects the brain's cytoarchitecture and assessed its test-retest reliability by using connectome fingerprints in Datasets 1 and 2. We also examined the VSN's behavioral relevance and further tested its predictive utility using connectome-based predictive modeling in Datasets 1 and 3. The VSN defined using the z-transformed interregional correlation showed significant spatial similarity with the cytoarchitectonic covariance network (rhos = 0.23 and 0.22 in Datasets 1 and 2, respectively; p < 0.01). The VSN also yielded high test-retest reliability, demonstrated by high identification accuracy (91% and 100% in Datasets 1 and 2, respectively). However, unlike the functional connectome (r > 0.31, p < 0.01), VSNs did not reliably predict individual differences in attention (r < 0.1, p > 0.3). This study demonstrates the biological validity and high reliability of the VSN to support brain fingerprinting of individual subjects, but not individual differences in attention.

Mitigating Long-Tailed Anomaly Score Distributions with Importance-Weighted Loss

arXiv (Cornell University) · 2026-01-05

Anomaly detection is crucial in industrial applications for identifying rare and unseen patterns to ensure system reliability. Traditional models, trained on a single class of normal data, struggle with real-world distributions where normal data exhibit diverse patterns, leading to class imbalance and long-tailed anomaly score distributions (LTD). This imbalance skews model training and degrades detection performance, especially for minority instances. To address this issue, we propose a novel importance-weighted loss designed specifically for anomaly detection. Compared to the previous method for LTD in classification, our method does not require prior knowledge of normal data classes. Instead, we introduce a weighted loss function that incorporates importance sampling to align the distribution of anomaly scores with a target Gaussian, ensuring a balanced representation of normal data. Extensive experiments on three benchmark image datasets and three real-world hyperspectral imaging datasets demonstrate the robustness of our approach in mitigating LTD-induced bias. Our method improves anomaly detection performance by 0.043, highlighting its effectiveness in real-world applications.

Abstract LB254: Redirecting cytomegalovirus immunity against breast tumors for immunotherapy

Cancer Research · 2026-04-17

Abstract Background: Breast tumors, particularly triple-negative cancers, are aggressive and frequently resistant to chemotherapy, with limited responses to immunotherapy. Despite high mutational burden and T-cell infiltration, clinical efficacy remains low. Based on evidence that neo/novel neoantigens elicit strong antitumor immunity, we hypothesize that systemic delivery of viral antigens combined with the tumor-targeting peptide iRGD can redirect pre-existing antiviral immunity to eliminate breast tumors. Methods: Cytomegalovirus (CMV), a β-herpesvirus, was chosen due to its high prevalence in the human population and its ability to elicit a robust and broadly reactive memory T cell response. Mice latently infected with murine CMV (MCMV) were orthotopically implanted with murine E0771 breast tumor cells in the mammary fat pad and subsequently treated with systemic administration of MCMV-derived T cell epitopes. Tumor progression was assessed biweekly via tumor size measurements using caliper, and immune cell infiltration and response were evaluated using histological analysis and flow cytometry. Statistical significance was determined using two-way ANOVA with Sidak’s post hoc correction, one-way ANOVA with Turkey’s post hoc correction and t-test. Results: Our new results demonstrate that MCMV-based therapy promotes the preferential accumulation of MCMV-specific T cells within breast tumors, leading to delayed tumor progression in mice infected and treated with MCMV-specific peptides compared with uninfected treated mice or infected mice treated with vehicle. The therapeutic efficacy was independent of iRGD co-administration. Notably, we observed heterogeneity in tumor growth control in the E0771 breast tumor model, allowing the identification of responder and non-responder groups. Immunophenotyping analyses revealed that tumor-infiltrating T cells displayed a highly activated phenotype, as well as enhanced cytotoxic potential, evidenced by elevated expression of granzymes A and B. These immune features correlated with increased tumor necrosis and enhanced T-cell infiltration within the tumor tissue of infected mice treated with MCMV-specific peptides. Conclusions: These findings demonstrate that CMV-specific memory T cells can be redirected to control breast tumors via systemic administration of CMV epitopes, and they were effective in curtailing tumor growth. Given that CMV infection is endemic and induces a huge memory T cell pool, this approach may have broad clinical applicability. Citation Format: Catarina Maia, Philip Salu, Rithika Medari, Remi Marrocco, Eduardo Lucero Meza, Kwangsun Yoo, Andrew Lowy, Christopher Benedict, Tatiana Hurtado de Mendoza. Redirecting cytomegalovirus immunity against breast tumors for immunotherapy [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 2 (Late-Breaking, Clinical Trial, and Invited Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(8_Suppl):Abstract nr LB254.

Semi-Supervised Hyperspectral Food Anomaly Detection Through Batch-Wise Distribution Alignment

2025-12-08

Metric-Driven Similarity Indices: Redefining Spectral Distance Comparisons in Hyperspectral Data

2025-02-18

Hyperspectral images (HSIs) offer rich spectral details but pose challenges in analyzing spectral vector distances due to high dimensionality and inter-class similarity. Existing distance metrics, while effective in specific cases, often fail to provide consistent comparisons across different tasks due to varying scales. This study proposes novel similarity score indices that normalize metrics onto a unified scale, ensuring fair, interpretable comparisons tailored to the unique properties of HSIs. Our evaluations on public datasets reveal the indices' ability to improve accuracy and reliability in spectral similarity assessments, addressing key challenges in HSI analysis.

Buffering brain aging: education moderates language impairment in Parkinson's disease

Frontiers in Cellular Neuroscience · 2025-08-20

Background: Cognitive reserve (CR) refers to the discrepancy between brain pathology and observed cognitive decline. While education is a key indicator of CR, its role as a potential moderator in the relationships between brain morphology and cognitive impairments in Parkinson's disease (PD) remains unclear. This study examined whether education affects the relationship between brain age and cognitive impairments in patients with PD. Methods: Data from 58 patients with PD were analyzed using a secondary dataset from the OpenNeuro database. Participants aged ≥55 years were on stable medications and underwent standardized neuropsychological assessments. Brain age predictions were generated from T1-weighted magnetic resonance imaging (MRI) using the brainageR package, and the brain age difference (BAD) was calculated after correction for regression dilution. The moderation effect of education on the relationship between BAD and cognition was assessed using Hayes' PROCESS macro. The primary outcome was cognitive performance across six domains: attention, executive function, language, learning and memory, visuospatial ability, and global cognition. Results: = 0.20). The relationship between BAD and language was steeper at lower education levels. No statistically significant moderation was found in the remaining five domains. Conclusion: Having more years of education is associated with buffering the effects of accelerated brain aging on language ability in PD.

Situating the salience and parietal memory networks in the context of multiple parallel distributed networks using precision functional mapping

Cell Reports · 2025-01-01 · 34 citations

Brain networks serving higher cognitive functions are widely distributed across frontal and posterior association zones. Two exceptions have been the parietal memory network (PMN) and salience network (SAL), which are typically restricted to posterior (e.g., posterior cingulate and lateral parietal cortex) and anterior (medial prefrontal and anterior insular cortex) areas, respectively. Using high-resolution neuroimaging, we show that individualized estimates of the PMN extend beyond the posterior set and encompass frontal and insula regions canonically ascribed to the SAL. This suggests that the SAL and PMN form a unified network: "SAL/PMN." Task-based analyses confirm that both anterior and posterior components of the SAL/PMN show recognition-related activity. Comparison of 3T and 7T data suggests that high-resolution data more readily revealed the unified network, underscoring the importance of fine-scale distinctions for veridical representation of brain networks. Importantly, the unified network better matches the expected parallel distributed network organization that is characteristic of association cortex.

Spectral Mixing Augmentation for Preventing False Positives from Hyperspectral Anomaly Detection

2025-08-18

In hyperspectral imaging, anomaly detection models often struggle with spectral mixing due to low spatial resolution or complex surface interactions. Existing models fail to generalize well across diverse spectral mixing scenarios, particularly near objects with irregular shapes or sizes. This leads to high false positive rates and increased inspection overhead, especially in industrial applications such as food quality assessment. To address this challenge, we propose a novel data augmentation framework that leverages a linear spectral mixing model to capture diverse real-world spectral compositions. The augmented data is integrated into the training pipeline of anomaly detection models to enhance their robustness. Extensive evaluations on real-world hyperspectral datasets demonstrate that our approach significantly reduces false positives while improving model reliability. The proposed framework provides a scalable and effective solution for industrial anomaly detection and quality control tasks.

Statistical Robustness in Anomaly Detection with Hard Rejection and Adaptive Weighting

2025-11-12

Deep learning models excel in anomaly detection by leveraging normal data to identify abnormalities. They conventionally assume a clean training dataset that consists of only normal classes due to the lack of abnormal training data. However, the inclusion of unintended anomalies in training data can degrade model performance, necessitating robust methods to handle such contamination. Aggressive rejection, which assumes a large amount of training data as potential anomalies, achieves high robustness. However, it does not fully exclude anomalies during the training process, and due to its fixed weight, it is challenging to balance performance between clean and contaminated data. In this paper, we propose statistical robustness in anomaly detection (SRAD), a novel statistical approach to enhance the robustness of aggressive rejection techniques. By employing the modified z-score to exclude high-confidence anomalies and leveraging kurtosis to adaptively adjust the weights for aggressive rejection, our method effectively addresses contamination in anomaly score distributions. Experiment results on two image datasets and thirty tabular datasets demonstrate that SRAD outperforms the state-of-the-art method by 0.039 AUROC, offering a stable and robust solution for real-world applications. Our approach underscores its potential for broad applicability across various domains requiring reliable anomaly detection.

GLESim: An Integrated New Similarity Index for Global and Local Spectral Analysis in Hyperspectral Imaging

2025-08-03

Hyperspectral imaging (HSI) facilitates comprehensive spectral analysis across multiple bands, unlocking significant potential in diverse industrial applications. Nevertheless, the complexities of hyperspectral data and the shortcomings of existing similarity metrics limit its practical utility. To overcome these limitations, we introduce GLESim, a novel similarity index combining three synergistic components: a RMSE-based term for capturing Global spectral trends, a nonlinear adjustment factor to enhance sensitivity to Local variations while suppressing outlier effects, and a Chebyshev-based term for detecting Extreme spectral deviations. Experimental validation on benchmark hyperspectral datasets demonstrates the superior performance of GLESim. These results position GLESim as a transformative tool for advancing HSI analysis methodologies, enabling precise spectral differentiation across varied applications.