 |
International Society of Science and Applied Technologies |
| GoogLenet with Squeeze-and-Excite Block Model for Bronchopulmonary Dysplasia Prediction | ||||
| Author |
Ya-Chi Hsu
|
|||
| Co-Author(s) |
Yu-Chieh Chen; Kuo-Ping Lin; Chih-Yuan Chu
|
|||
| Abstract | Early diagnosis of bronchopulmonary dysplasia from chest X-ray (CXR) images remains challenging due to limited annotated data, image variability, and overlapping radiographic features. To address these issues, this study proposes a deep learning framework integrating GoogLeNet with a Squeeze-and-Excitation (SE) block to enhance feature representation. The model combines multi-scale feature extraction with channel-wise attention to emphasize clinically relevant information. Experiments were conducted on a neonatal CXR dataset (100 images) from a medical center in central Taiwan. Model performance was evaluated using Accuracy, Precision, Recall, and F1-score across 30 independent runs. Results show that the proposed GoogLeNet+SE-block achieves the best performance, with an accuracy of 0.817±0.10 and recall of 0.751, outperforming GoogLeNet, U-Net, and ResNet. These findings demonstrate that integrating attention mechanisms significantly improves robustness and diagnostic sensitivity under data-constrained conditions. The proposed framework has strong potential for clinical decision support applications in bronchopulmonary dysplasia imaging.
|
|||
| Keywords | Bronchopulmonary dysplasia; Chest X-ray (CXR); Deep learning; GoogLeNet; Squeeze-and-Excitation (SE) block | |||
| Article #: RQD2026-45 | ||||
Proceedings of 31st ISSAT International Conference on Reliability & Quality in Design |