Preliminary exploration on using entropy-weighted hybrid pooling in CNN for ultrasound breast cancer detection - Summary - MDSpire

Preliminary exploration on using entropy-weighted hybrid pooling in CNN for ultrasound breast cancer detection

  • By

  • Ratapong Onjun

  • Papon Tantiwanichanon

  • Songkiat Lowmunkhong

  • Tanakorn Sritarapipat

  • Sayan Kaennakham

  • Niwatchai Namwichaisirikul

  • Kitirat Phattaramarut

  • July 8, 2026

  • 0 min

Share

Objective:

To evaluate an adaptive entropy-weighted hybrid pooling approach in CNNs for improving ultrasound breast cancer detection.

Approach:
  • Methodology: The study introduced a hybrid pooling method that combines Max and Average pooling based on local image complexity measured by Shannon entropy. Two CNN architectures (3-block and 4-block) were tested on a dataset of 9,016 ultrasound images.
Key Findings:
  • The 3-block CNN with hybrid pooling achieved an accuracy of 93.98% ± 1.72% (AUC = 0.9870), outperforming max pooling at 92.72% ± 0.85% (AUC = 0.9815).
  • The 4-block CNN with hybrid pooling showed competitive accuracy at 92.90% (AUC not reported) compared to max pooling at 94.79%.
Interpretation:

Limitations:
  • The study is preliminary and requires additional validation and clinical integration.
  • Results from the 4-block CNN were based on a single run, limiting robustness.
Conclusion:

Original Source(s)

Preliminary exploration on using entropy-weighted hybrid pooling in CNN for ultrasound breast cancer detection

  • by Ratapong Onjun, Papon Tantiwanichanon, Songkiat Lowmunkhong, Tanakorn Sritarapipat, Sayan Kaennakham, Niwatchai Namwichaisirikul, Kitirat Phattaramarut

  • July 8, 2026

Related Content

Multi-scale information bottleneck with confidence-weighted decision fusion for robust breast ultrasound lesion classification

The Unrealized Potential of Cascade Testing

Subtype-specific sonographic signatures and clinicopathological features of metaplastic breast cancer: a 50-case cohort study