arXiv subjects

Anastasiia Studenikina

Publications and source records attributed to Anastasiia Studenikina.

3 recordsLinked to original sources

arXiv2026

Deep Learning for Automated Quantification of Tumor-Associated Macrophages from H&E-Stained Slides in Diffuse Large B-Cell Lymphoma

While M2-polarized tumor-associated macrophages (TAMs) have been established as indicators of disease aggressiveness in diffuse large B-cell lymphoma (DLBCL), traditional CD163 immunohistochemistry (IHC) remains resource-intensive. This study aims to investigate the feasibility of using deep learning to quantify TAMs directly from standard hematoxylin and eosin-stained (HE) tissue sections. Using a curated dataset of 52 patients with DLBCL, with high-resolution HE images and IHC-validated annotations (1,713 TAM instances), five architectures were evaluated: U-Net, Swin-U-Net, Cerberus-U-Net3+, YOLOv11, and HoVer-Net. High CD163 TAM density (>20.04%) was associated with significantly reduced overall survival (HR 2.73; 95% CI:1.21-6.16; p=0.012) and progression-free survival (HR 2.88; 95% CI: 1.23-6.76; p=0.011). In multivariate Cox proportional hazards analysis adjusting for IPI, molecular subtype (GCB/non-GCB per Hans algorithm), EBV status, and age, CD163 TAM density showed a prognostic trend for overall survival (HR 3.24; 95% CI: 0.94-11.15; p=0.062) and progression-free survival (HR 2.41; 95% CI: 0.80-7.32; p=0.120). Among the evaluated models, the domain-specific Cerberus-U-Net3+ achieved the highest sensitivity (Recall 0.656), while the Transformer-based Swin-U-Net demonstrated superior segmentation fidelity (Precision 0.694, F1-score 0.633). Additionally, the survival analysis based on the predicted Swin-U-Net CD163 level revealed a 20.8% cutoff point for patients with high and low CD163 levels near IHC, as well as a downward trend in overall survival among patients with higher predicted CD163 values of 2.63 (95% CI: 0.85-8.33; p=0.083). These findings suggest that deep learning architectures using shifted-window self-attention show potential as a candidate surrogate for IHC that is scalable and cost-effective for prognostic assessment of TAMs in DLBCL.

eess.IV↗

arXiv2026

GCond: Gradient Conflict Resolution via Accumulation-based Stabilization for Large-Scale Multi-Task Learning

In multi-task learning (MTL), gradient conflict poses a significant challenge. Effective methods for addressing this problem, including PCGrad, CAGrad, and GradNorm, in their original implementations are computationally demanding, which significantly limits their application in modern large models such as transformers. We propose Gradient Conductor (GCond), a method that builds upon PCGrad principles by combining them with gradient accumulation and an adaptive arbitration mechanism. We evaluated GCond on self-supervised multi-task learning tasks using MobileNetV3-Small and ConvNeXt architectures on the ImageNet 1K dataset and a combined head and neck CT scan dataset, comparing the proposed method against baseline linear combinations and state-of-the-art gradient conflict resolution methods. The classical and stochastic approaches of GCond were analyzed. The stochastic mode of GCond achieved a two-fold computational speedup while maintaining optimization quality, and demonstrated superior performance across all evaluated metrics, achieving lower L1 and SSIM losses compared to other methods on both datasets, and demonstrating superior generalization in heterogeneous scenarios: GCond improved ImageNet Top-1 Accuracy by 4.5% over baselines and prevented confidence overfitting in medical diagnosis tasks. GCond exhibited high scalability, being successfully applied to both compact models: MobileNetV3-Small and ConvNeXt-tiny; and large architecture ConvNeXtV2-Base. It also showed compatibility with modern optimizers such as AdamW and Lion/LARS. Therefore, GCond offers a scalable and efficient solution to the problem of gradient conflicts in multi-task learning.

cs.LG↗

arXiv2025

BertsWin: Resolving Topological Sparsity in 3D Masked Autoencoders via Component-Balanced Structural Optimization

The application of self-supervised learning (SSL) and Vision Transformers (ViTs) approaches demonstrates promising results in the field of 2D medical imaging, but the use of these methods on 3D volumetric images is fraught with difficulties. Standard Masked Autoencoders (MAE), which are state-of-the-art solution for 2D, have a hard time capturing three-dimensional spatial relationships, especially when 75% of tokens are discarded during pre-training. We propose BertsWin, a hybrid architecture combining full BERT-style token masking using Swin Transformer windows, to enhance spatial context learning in 3D during SSL pre-training. Unlike the classic MAE, which processes only visible areas, BertsWin introduces a complete 3D grid of tokens (masked and visible), preserving the spatial topology. And to smooth out the quadratic complexity of ViT, single-level local Swin windows are used. We introduce a structural priority loss function and evaluate the results of cone beam computed tomography of the temporomandibular joints. The subsequent assessment includes TMJ segmentation on 3D CT scans. We demonstrate that the BertsWin architecture, by maintaining a complete three-dimensional spatial topology, inherently accelerates semantic convergence by a factor of 5.8x compared to standard ViT-MAE baselines. Furthermore, when coupled with our proposed GradientConductor optimizer, the full BertsWin framework achieves a 15-fold reduction in training epochs (44 vs 660) required to reach state-of-the-art reconstruction fidelity. Analysis reveals that BertsWin achieves this acceleration without the computational penalty typically associated with dense volumetric processing. At canonical input resolutions, the architecture maintains theoretical FLOP parity with sparse ViT baselines, resulting in a significant net reduction in total computational resources due to faster convergence.

cs.CV↗