Abstract
Manual analysis of positive and negative in H&E-stained histopathology images is critical yet remains subjective and time-consuming in breast cancer prognosis. This study proposes an automated multi-class segmentation approach using a U-Net architecture to distinguish between these cell types. We evaluated the method across six experimental scenarios, comparing two data preparation strategies (image resizing vs. image tiling) and multiple hyperparameter configurations, including learning rate and regularization techniques. Results show that image tiling, which preserves high-resolution details, significantly outperforms resizing. The model with the highest performance obtained a Mean Intersection over Union (IoU). 0.6619 and Dice Coefficient 0.8664 on the test set by combining image tiling with using reduced learning rate of 0.00005 and stronger regularization (Dropout 0.3 and L2). Despite limitations such as class imbalance and the absence of data augmentation, this study highlights the proposed method's potential as a precise and efficient tool for automated Ki-67 analysis in clinical settings.
| Original language | English |
|---|---|
| Title of host publication | Proceedings of 2025 IEEE International Conference on Internet of Things and Intelligence Systems, IoTaIS 2025 |
| Publisher | Institute of Electrical and Electronics Engineers Inc. |
| Pages | 171-176 |
| Number of pages | 6 |
| ISBN (Electronic) | 9798331581244 |
| DOIs | |
| Publication status | Published - 2025 |
| Event | 2025 IEEE International Conference on Internet of Things and Intelligence Systems, IoTaIS 2025 - Hybrid, Bali, Indonesia Duration: 6 Nov 2025 → 8 Nov 2025 |
Publication series
| Name | Proceedings of 2025 IEEE International Conference on Internet of Things and Intelligence Systems, IoTaIS 2025 |
|---|
Conference
| Conference | 2025 IEEE International Conference on Internet of Things and Intelligence Systems, IoTaIS 2025 |
|---|---|
| Country/Territory | Indonesia |
| City | Hybrid, Bali |
| Period | 6/11/25 → 8/11/25 |
UN SDGs
This output contributes to the following UN Sustainable Development Goals (SDGs)
-
SDG 3 Good Health and Well-being
-
SDG 7 Affordable and Clean Energy
Keywords
- Breast Cancer
- Deep Learning
- Hematoxylin Eosin
- Multi-Class Segmentation
- U-Net
Fingerprint
Dive into the research topics of 'Multi-Class Cell Segmentation in H&E Images Using Deep Learning'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver