Search bioRxiv⌕ Search
Home/Archive/bioRxiv

bioRxiv · 10.64898/2026.10.29.762804

NeuroWeave-3D: Uncertainty-Aware Evidence Completion for Brain Tumor Segmentation with Missing MRI Modalities

Miller, D.·Weber, L. A.
View original source ↗Export metadata ↓

Abstract

Multi-parametric MRI segmentation models are accurate when all prescribed sequences are available but can be brittle under missing or corrupted modalities. We propose NeuroWeave-3D, a 3D glioma segmentation framework that treats missing modalities as uncertain evidence rather than deterministic channels to reconstruct. The model combines spatial modality routing, probabilistic latent feature completion, complete-to-missing consistency, and voxel-level uncertainty estimation. A curriculum over all non-empty MRI modality subsets trains one network for arbitrary sequence availability. We define a BraTS-centered evaluation that reports complete-input accuracy, macro and worst-case missing-modality performance, corruption robustness, calibration, and failure detection.

Explore related subjects

bioinformaticsPreprint

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Miller, D., Weber, L. A.. 2026-11-04. NeuroWeave-3D: Uncertainty-Aware Evidence Completion for Brain Tumor Segmentation with Missing MRI Modalities. https://doi.org/10.64898/2026.10.29.762804

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Cellular composition and the interpretation of candidate RNA readouts of trogocytosis in colorectal cancer

Background: RNA measurements of genes associated with trogocytosis may reflect cellular composition rather than contact-dependent membrane acquisition. We evaluated an eight-gene candidate module in colorectal cancer (CRC) using expression- and lineage-matched random modules, independent compartment analysis and patient-level inference. A separate rare-transcript definition and cancer-model fitness screens provided complementary analyses. Results: In 62 GSE178341 patients, the module correlated with T-cell fraction ({rho}=0.767), but this association was typical of 10,000 matched modules (null median 0.766; one-sided empirical P=0.493; equal-tail two-sided P=0.985). A post hoc closest-10% matching sensitivity gave P=0.571. In 23 independent GSE132465 patients, the tumour-wide correlation was 0.735 (permutation P=0.00060) and the epithelial correlation was 0.376 (P=0.0892). The difference in absolute correlations was 0.359, with a 95% confidence interval of -0.041 to 0.759, leaving attenuation uncertain. Among 43,817 leukocytes, 16 CEACAM5-positive candidates arose from four patients; 15 had expression [≤]2 transcripts per million. Candidate counts were threshold-sensitive. No survival association was detected for the module in two cohorts. SCD fitness dependency reproduced across CRC organoids and cell lines, whereas adjusted SCD-VPS72 and exploratory ATF3-SMYD2 co-dependencies did not. Exploratory CH25H, ATF3 and FADS2 analyses yielded no Hallmark enrichment meeting the multiple-testing threshold. Conclusions: Matched-module calibration placed the strong association with T-cell abundance within the range expected for comparable immune-lineage genes. Independent compartment analysis was consistent with a contribution from cellular composition, although attenuation remained uncertain. These results identify the comparison conditions needed to interpret the tested RNA readouts. Linking them to membrane transfer requires event-resolved protein measurements.

bioinformatics↗
bioRxiv2026

Differential Gene Expression and Functional Enrichment Analysis of Breast Tumor and Normal Tissue RNA-seq Data: A Reanalysis of GEO Dataset GSE210787

Background: RNA sequencing (RNA-seq) allows a global assessment of expression differences between biological states. The present study reanalyzed publicly available RNA-seq data from human breast tumors and normal tissues to identify DEGs and to investigate functional enrichment. Methods: Twelve paired-end RNA-seq samples from the GEO series GSE210787 were analyzed, including six tumor and six normal samples. Reads were obtained using the SRA Toolkit, checked with FastQC and MultiQC, trimmed with fastp, aligned to the GRCh38.p14 reference with STAR, and quantified at the gene level with featureCounts. Differential expression analysis was performed using DESeq2 in the Galaxy environment. The project report details criteria of adjusted p-value < 0.05 and absolute log2 fold change [&ge;] 1. Gene Ontology enrichment was performed using g:Profiler and KEGG overrepresentation analysis was performed separately for the up- and down-regulated gene sets. Results: The project report contains a list of 120 differentially expressed genes (DEGs), including 69 upregulated and 51 downregulated genes in tumor tissue compared with normal tissue. One upregulated-gene GO Cellular Component term, cell periphery (GO:0071944), was observed with a P-value of approximately 0.017585 and 29 overlapping genes. The original report does not specify if this P-value is nominal or adjusted. No significant GO terms were found for the down-regulated gene set. After multiple testing corrections, no KEGG pathway was significant; the lowest adjusted p-value for the upregulated group was about 0.71393. Conclusion: The project identified differences in expression between tumor and normal samples, but no KEGG pathway reached the set FDR significance threshold. The final interpretation needs to check sample metadata, DESeq2 contrast, GO significance statistic, KEGG background universe and analytical parameters against the original results. Keywords: RNA-seq; breast tumor; differential expression; DESeq2; Gene Ontology; KEGG pathway; transcriptomics; cell periphery.

bioinformatics↗
bioRxiv2026

A trustworthy semantic knowledge network linking data-driven cell phenotypes to diseases and drugs

Genome-wide association and other genetic studies have identified genes involved in disease pathogenesis, and their encoded proteins have provided therapeutic targets. However, little is known about which cells are the functional home of these disease-associated genes and proteins. Single-cell technologies are enabling high-resolution profiling of cell phenotypes and revealing cellular contexts of disease, yet much of the resulting knowledge remains embedded in individual datasets and scientific literature. To address these gaps, the National Library of Medicine has developed Cell Knowledge Network (NLM-CKN, https://nlm-ckn.org). Reproducible computational workflows have generated cell phenotype knowledge for 2,585 cell sets and 5,139 selectively expressed genes from 88 datasets covering 10 human organs. Mapping of these cell sets to cell types grounds Cell Ontology terms in experimental evidence. This experimental knowledge is integrated with external knowledge of anatomical and cell-type hierarchies, gene-disease associations, and therapeutic interventions. A semantic schema defines node types and relations, enabling knowledge to be transformed into triple assertions and loaded into an ArangoDB database with provenance and quantitative evidence. The NLM-CKN interface supports graphical exploration of the cellular constituents of healthy human tissues. Finally, topological analyses generate insights into cellular contributions to disease pathogenesis, cellular contexts for therapeutic intervention, and drug-repurposing hypotheses.

bioinformatics↗