[공지사항]을 빙자한 안부와 근황 
Show more

Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Site-Level Fine-Tuning with Progressive Layer Freezing: Towards Robust Prediction of Bronchopulmonary Dysplasia from Day-1 Chest Radiographs in Extremely Preterm Infants

Created by
  • Haebom

Author

Sybelle Goedicke-Fritz (Department of General Pediatrics and Neonatology, Saarland University, Campus Homburg, Homburg/Saar, Germany), Michelle Bous (Department of General Pediatrics and Neonatology, Saarland University, Campus Homburg, Homburg/Saar, Germany), Annika Engel (Chair for Clinical Bioinformatics, Saarland Informatics Campus, Saarland University, Saarbr ucken, Germany), Matthias Flotho (Chair for Clinical Bioinformatics, Saarland Informatics Campus, Saarland University, Saarbr ucken, Germany, Helmholtz Institute for Pharmaceutical Research Saarland), Pascal Hirsch (Chair for Clinical Bioinformatics, Saarland Informatics Campus, Saarland University, Saarbr ucken, Germany), Hannah Wittig (Department of General Pediatrics and Neonatology, Saarland University, Campus Homburg, Homburg/Saar, Germany), Dino Milanovic (Chair for Clinical Bioinformatics, Saarland Informatics Campus, Saarland University, Saarbr ucken, Germany), Dominik Mohr (Department of General Pediatrics and Neonatology, Saarland University, Campus Homburg, Homburg/Saar, Germany), Mathias Kaspar (Digital Medicine, University Hospital of Augsburg, Augsburg, Germany), Sogand Nemat (Department of Radiology, and Interventional Radiology, University Hospital of Saarland, Homburg, Germany), Dorothea Kerner (Department of Radiology, and Interventional Radiology, University Hospital of Saarland, Homburg, Germany), Arno B ucker (Department of Radiology, and Interventional Radiology, University Hospital of Saarland, Homburg, Germany), Andreas Keller (Chair for Clinical Bioinformatics, Saarland Informatics Campus, Saarland University, Saarbrucken , Germany, Helmholtz Institute for Pharmaceutical Research Saarland, Pharma Science Hub), Sascha Meyer (Clinical Center Karlsruhe, Franz-Lust Clinic for Paediatrics, Karlsruhe, Germany), Michael Zemlin (Department of General Pediatrics and Neonatology, Saarland University, Campus Homburg, Homburg/Saar, Germany), Philipp Flotho (Chair for Clinical Bioinformatics, Saarland Informatics Campus, Saarland University, Saarbrücken , Germany, Helmholtz Institute for Pharmaceutical Research Saarland)

Outline

Herein, we present a deep learning-based approach for early prognosis and outcome prediction of bronchopulmonary dysplasia (BPD), a chronic lung disease that occurs in 35% of extremely low birth weight infants. We fine-tune a ResNet-50 model pretrained on adult chest ___T9998_____x-rays using 163 extremely low birth weight infants (gestational age ≤32 weeks, body weight 401-999 g) taken within the first 24 hours after birth. We use progressive layer freezing and discriminative learning rate to prevent overfitting, and apply CutMix augmentation and linear probing. The best-performing model achieves an AUROC of 0.78 ± 0.10, a balanced accuracy of 0.69 ± 0.10, and an F1-score of 0.67 ± 0.11 for predicting moderate/severe BPD outcomes. We found that domain-specific pre-training outperformed ImageNet initialization (p = 0.031). Routine IRDS ratings had limited prognostic value (AUROC 0.57 ± 0.11), confirming the necessity of learned markers. Progressive freezing and linear probing make it a computationally efficient method suitable for field-level implementation and future federated learning deployments.

Takeaways, Limitations

•
Takeaways:
◦
Development of a deep learning model that can accurately predict BPD outcomes using chest X radiographs taken within 24 hours of birth.
◦
Confirms the importance of domain-specific pre-training.
◦
Computationally efficient model implementation via progressive hierarchical freezing and linear probing.
◦
Presenting field-level implementation and deployment possibilities for federated learning.
◦
More accurate BPD prediction than existing IRDS grades.
•
Limitations:
◦
Using a relatively small dataset (163 people).
◦
Further validation of the model's generalization performance is needed.
◦
It is not a perfect prediction as AUROC, balanced accuracy, and F1-score all do not reach 1.
◦
Further research is needed to determine generalizability across different racial and ethnic groups.
👍
Made with Slashpage