arXiv subjects

Toru Nagasaka

Publications and source records attributed to Toru Nagasaka.

4 recordsLinked to original sources

arXiv2026

Revisiting the Role of Foundation Models in Cell-Level Histopathological Image Analysis under Small-Patch Constraints -- Effects of Training Data Scale and Blur Perturbations on CNNs and Vision Transformers

Background and objective: Cell-level pathological image analysis requires working with extremely small image patches (40x40 pixels), far below standard ImageNet resolutions. It remains unclear whether modern deep learning architectures and foundation models can learn robust and scalable representations under this constraint. We systematically evaluated architectural suitability and data-scale effects for small-patch cell classification. Methods: We analyzed 303 colorectal cancer specimens with CD103/CD8 immunostaining, generating 185,432 annotated cell images. Eight task-specific architectures were trained from scratch at multiple data scales (FlagLimit: 256--16,384 samples per class), and three foundation models were evaluated via linear probing and fine-tuning after resizing inputs to 224x224 pixels. Robustness to blur was assessed using pre- and post-resize Gaussian perturbations. Results: Task-specific models improved consistently with increasing data scale, whereas foundation models saturated at moderate sample sizes. A Vision Transformer optimized for small patches (CustomViT) achieved the highest accuracy, outperforming all foundation models with substantially lower inference cost. Blur robustness was comparable across architectures, with no qualitative advantage observed for foundation models. Conclusion: For cell-level classification under extreme spatial constraints, task-specific architectures are more effective and efficient than foundation models once sufficient training data are available. Higher clean accuracy does not imply superior robustness, and large pre-trained models offer limited benefit in the small-patch regime.

cs.CV↗

arXiv2025

Spatially-extended Flow Phixer (SpeF-Phixer): A Spatially Extended $φ$-Mixing Framework for Gene Regulatory Causal Inference in Spatial Gene Field

Background and objective: Spatial transcriptomics provides rich spatial context but lacks sufficient resolution for large-scale causal inference. We developed SpeF-Phixer, a spatially extended phi-mixing framework integrating whole-slide image (WSI)-derived spatial cell distributions with mapped scRNA-seq expression fields to infer directed gene regulatory triplets with spatial coherence. Methods: Using CD103/CD8-immunostained colorectal cancer WSIs and publicly available scRNA-seq datasets, spatial gene fields were constructed around mapped cells and discretized for signed phi-mixing computation. Pairwise dependencies, directional signs, and triplet structures were evaluated through kNN-based neighborhood screening and bootstrap consensus inference. Mediation and convergence were distinguished using generalized additive models (GAMs), with spatial validity assessed by real-null comparisons and database-backed direction checks. Results: Across tissue patches, the pipeline reduced approximately 3.6x10^4 triplet candidates to a reproducible consensus set (approximately 3x10^2 per patch). The downstream edge (Y to Z) showed significant directional bias consistent with curated regulatory databases. Spatial path tracing demonstrated markedly higher coherence for real triplets than for null controls, indicating that inferred chains represent biologically instantiated regulatory flows. Conclusion: SpeF-Phixer extracts spatially coherent, directionally consistent gene regulatory triplets from histological images. This framework bridges single-cell molecular profiles with microenvironmental organization and provides a scalable foundation for constructing spatially informed causal gene networks.

q-bio.QM↗

arXiv2025

Reliability Assessment Framework Based on Feature Separability for Pathological Cell Image Classification under Prior Bias

Background and objective: Prior probability shift between training and deployment datasets challenges deep learning-based medical image classification. Standard correction methods reweight posterior probabilities to adjust prior bias, yet their benefit is inconsistent. We developed a reliability framework identifying when prior correction helps or harms performance in pathological cell image analysis. Methods: We analyzed 303 colorectal cancer specimens with CD103/CD8 immunostaining, yielding 185,432 annotated cell images across 16 cell types. ResNet models were trained under varying bias ratios (1.1-20$\times$). Feature separability was quantified using cosine similarity-based likelihood quality scores, reflecting intra- versus inter-class distinctions in learned feature spaces. Multiple linear regression, ANOVA, and generalized additive models (GAMs) evaluated associations among feature separability, prior bias, sample adequacy, and F1 performance. Results: Feature separability dominated performance ($\beta = 1.650$, $p < 0.001$), showing 412-fold stronger impact than prior bias ($\beta = 0.004$, $p = 0.018$). GAM analysis showed strong predictive power ($R^2 = 0.876$) with mostly linear trends. A quality threshold of 0.294 effectively identified cases requiring correction (AUC = 0.610). Cell types scoring $>0.5$ were robust without correction, whereas those $<0.3$ consistently required adjustment. Conclusion: Feature extraction quality, not bias magnitude, governs correction benefit. The proposed framework provides quantitative guidance for selective correction, enabling efficient deployment and reliable diagnostic AI.

q-bio.QM↗

arXiv2024

Novel Methods for Analyzing Cellular Interactions in Deep Learning-Based Image Cytometry: Spatial Interaction Potential and Co-Localization Index

The study presents a novel approach for quantifying cellular interactions in digital pathology using deep learning-based image cytometry. Traditional methods struggle with the diversity and heterogeneity of cells within tissues. To address this, we introduce the Spatial Interaction Potential (SIP) and the Co-Localization Index (CLI), leveraging deep learning classification probabilities. SIP assesses the potential for cell-to-cell interactions, similar to an electric field, while CLI incorporates distances between cells, accounting for dynamic cell movements. Our approach enhances traditional methods, providing a more sophisticated analysis of cellular interactions. We validate SIP and CLI through simulations and apply them to colorectal cancer specimens, demonstrating strong correlations with actual biological data. This innovative method offers significant improvements in understanding cellular interactions and has potential applications in various fields of digital pathology.

q-bio.QM↗