EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Enhancing Histopathology Images with Multi-Level Deep Learning Approaches

Nalavadi Srikantha, Chandrashekhar K

Computer and Decision Making: An International Journal · 09.11.2026 · 3 (8S) · 943-977

DOI: 10.1922/comdem.v3i8s.19239 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Computer and Decision Making: An International Journal, online erschienen am 09.11.2026. Verfasst von Nalavadi Srikantha und 1 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate reading of histopathology images is vital for disease assessment. Digital slides often suffer from restricted scanner resolution and image loss. Such limitations can obscure critical cellular features during clinical evaluation. This study compares modern image enhancement methods for digital pathology improvement. Four Super Resolution models were selected: PNet-U, Real-ESRGAN, SwinIR, and MIRNet-v2. Each model applies deep learning to enhance Low-Resolution medical imagery. The evaluation combined numerical analysis with perceptual quality inspection. Metrics used included PSNR, SSIM and expert-based visual grading. PNet-U produced the best quantitative accuracy across tested lung image datasets. It achieved 37.81 PSNR and 0.96 SSIM at 2× upscaling. SwinIR and Real-ESRGAN showed slightly lower values but richer perceptual detail retention. Its images appeared visually sharper and closer to original pathology textures. The SwinIR and Real-ESRGAN models achieved average PSNR values of 35.81 dB and 33.53 dB, respectively, with corresponding SSIM values of 0.95 and 0.92. MIRNet-v2 produced stable but comparatively lower reconstruction performance. It recorded 32.90 PSNR and 0.95 SSIM on lung images. For colon and PR-IHC tissues, similar moderate trends were observed. Real-ESRGAN achieved stronger realism despite minor metric differences. This indicates its robustness in complex histological reconstructions. A balance between visual quality and numerical accuracy remains essential. The findings suggest PNet-U as the most practical enhancement option. Its integration could improve diagnostic workflows and strengthen clinical interpretation reliability.

Zitieren

APA

Srikantha, N., & K, C. (2026). Enhancing Histopathology Images with Multi-Level Deep Learning Approaches. Computer and Decision Making: An International Journal, 3(8S), 943-977. https://doi.org/10.1922/comdem.v3i8s.19239

Vancouver

Srikantha N, K C. Enhancing Histopathology Images with Multi-Level Deep Learning Approaches. Computer and Decision Making: An International Journal. 2026;3(8S):943-977. doi:10.1922/comdem.v3i8s.19239

BibTeX
@article{Srikantha2026,
  author = {Srikantha, Nalavadi and K, Chandrashekhar},
  title = {Enhancing Histopathology Images with Multi-Level Deep Learning Approaches},
  journal = {Computer and Decision Making: An International Journal},
  year = {2026},
  volume = {3},
  number = {8S},
  pages = {943--977},
  doi = {10.1922/comdem.v3i8s.19239}
}
Kostenlos starten →