EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Comparative Evaluation of Deep CNN Architectures for Pattern Recognition in Mouse Brain Tissue Section Images

Zhong Hu, Partha S. Saha, Wenfeng An

Bioengineering · 07.11.2026 · 13 (11) · 1320

DOI: 10.3390/bioengineering13111320 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Bioengineering, online erschienen am 07.11.2026. Verfasst von Zhong Hu und 2 weiteren Autor:innen. Die Arbeit nennt 72 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate pattern recognition in biological cellular images is essential for quantitative biomedical analysis, yet remains challenging due to substantial variability in cellular morphology, tissue architecture, imaging noise, and staining variability. Conventional image processing approaches often rely on handcrafted features and fixed segmentation criteria, resulting in limited robustness and reduced generalization across divers imaging conditions. In this study, using atlas-guided annotations from adult mouse brain tissue sections, a curated image dataset representing six anatomically distinct brain regions was established for training and evaluation. Four representative deep convolutional neural network (CNN) architectures, i.e., AlexNet and ResNets (ResNet-18, ResNet-50, and Res-Net-101), were systematically evaluated within a unified MATLAB-based (R2025b) transfer-learning framework. The proposed approach combines CNN embeddings with color, texture, multi-feature similarity, and intensity descriptors to improve feature representation in heterogeneous biological environments. The modeling results were evaluated by multiple metrics, such as Per-class precision, recall (sensitivity), F1-score, and ablation study. Results demonstrate a clear depth–performance relationship, with ResNet-50 and ResNet-101 achieving superior accuracy (exceeding 90%), robustness, and generalization compared with AlexNet and ResNet-18. While deeper networks require higher computational resources. Comparative evaluation of multiple CNN architectures further illustrates the effectiveness of deep residual learning for atlas-guided brain tissue pattern recognition. The proposed framework provides a practical computational pipeline for biological image analysis and may support future applications in neuroscience, digital pathology, and automated tissue characterization.

Zitieren

APA

Hu, Z., Saha, P. S., & An, W. (2026). Comparative Evaluation of Deep CNN Architectures for Pattern Recognition in Mouse Brain Tissue Section Images. Bioengineering, 13(11), 1320. https://doi.org/10.3390/bioengineering13111320

Vancouver

Hu Z, Saha PS, An W. Comparative Evaluation of Deep CNN Architectures for Pattern Recognition in Mouse Brain Tissue Section Images. Bioengineering. 2026;13(11):1320. doi:10.3390/bioengineering13111320

BibTeX
@article{Hu2026,
  author = {Hu, Zhong and Saha, Partha S. and An, Wenfeng},
  title = {Comparative Evaluation of Deep CNN Architectures for Pattern Recognition in Mouse Brain Tissue Section Images},
  journal = {Bioengineering},
  year = {2026},
  volume = {13},
  number = {11},
  pages = {1320},
  doi = {10.3390/bioengineering13111320}
}
Kostenlos starten →