EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Beyond Size and Class Balance: Alpha as a New Dataset Quality Metric for Deep Learning

Josiah D Couch iD, Rima Arnaout iD, Ramy Arnaout iD

Bioengineering · 24.07.2026 · 13 (8) · 854

DOI: 10.3390/bioengineering13080854 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Bioengineering, online erschienen am 24.07.2026. Verfasst von Josiah D Couch und 2 weiteren Autor:innen. Beteiligte Einrichtung: Department of Pathology, Beth Israel Deaconess Medical Center (BIDMC), Boston, MA 02215, USA. Die Arbeit nennt 60 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: NIAID NIH HHS, NHLBI NIH HHS, Massachusetts Life Sciences Center.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

In deep-learning-based image classification, achieving high performance requires diverse training sets. However, the current best practice—maximizing dataset size and class balance—does not guarantee dataset diversity. We hypothesized that, for a given model architecture, performance improves by maximizing diversity more directly. To test this hypothesis, we introduce a comprehensive framework of “sentropic” diversity measures or “sentropies” from ecology that generalizes familiar quantities like Shannon entropy by accounting for similarities among images. Size and class balance emerge as special cases. Analyzing thousands of subsets from seven medical-imaging datasets showed that the best correlates of performance were not size or class balance but A—“big alpha”—a set of sentropies interpreted as the effective number of image–class pairs in the dataset, after accounting for similarities among images. One of these, A0, explained 67% of the variance in balanced accuracy, vs. 54% for class balance and just 39% for size. The best pair of measures was size-plus-A1 (79%), which outperformed size-plus-class-balance (74%). Subsets with the largest A0 performed up to 16% better than those with the largest size (median improvement, 8%). We propose developing methods to maximize A as a way to improve deep learning performance in medical imaging.

Schlagwörter

Stichwörter der Autor:innen

artificial neural networkscomputational cardiologycomputational pathologycomputer visiondataset qualitydiversityentropymedical imagingsimilarity

Autor:innen und Institutionen

  • Department of Pathology, Beth Israel Deaconess Medical Center (BIDMC), Boston, MA 02215, USA.
  • Department of Medicine, University of California San Francisco, San Francisco, CA 94143, USA.
  • Bakar Institute for Computational Health Sciences, University of California San Francisco, San Francisco, CA 94143, USA.
  • Center for Intelligent Imaging, University of California San Francisco, San Francisco, CA 94143, USA.
  • Division of Clinical Informatics, Department of Medicine, Beth Israel Deaconess Medical Center (BIDMC), Boston, MA 02215, USA.
  • Harvard Medical School, Harvard University, Boston, MA 02115, USA.

ORCID-Profile: Josiah D Couch, Rima Arnaout, Ramy Arnaout

Förderung

  • NIAID NIH HHS
  • NHLBI NIH HHS
  • Massachusetts Life Sciences Center
  • National Institutes of Health
  • Gordon and Betty Moore Foundation
  • Tianqiao and Chrissy Chen Institute
  • Food and Drug Administration

Zitieren

APA

Couch, J. D., Arnaout, R., & Arnaout, R. (2026). Beyond Size and Class Balance: Alpha as a New Dataset Quality Metric for Deep Learning. Bioengineering, 13(8), 854. https://doi.org/10.3390/bioengineering13080854

Vancouver

Couch JD, Arnaout R, Arnaout R. Beyond Size and Class Balance: Alpha as a New Dataset Quality Metric for Deep Learning. Bioengineering. 2026;13(8):854. doi:10.3390/bioengineering13080854

BibTeX
@article{Couch2026,
  author = {Couch, Josiah D and Arnaout, Rima and Arnaout, Ramy},
  title = {Beyond Size and Class Balance: Alpha as a New Dataset Quality Metric for Deep Learning},
  journal = {Bioengineering},
  year = {2026},
  volume = {13},
  number = {8},
  pages = {854},
  doi = {10.3390/bioengineering13080854}
}
Kostenlos starten →