Full name: Nguyen Thi Tan Tien
Officical thesis title: RESEARCH AND DEVELOPMENT OF DEEP LEARNING MODELS FOR CHEST X-RAY-BASED DIAGNOSIS OF OCCUPATIONAL SILICOSIS
Major: Computer Science Code: 9 48 01 01
Supervisors:
1: Assoc. Prof. Dr. Pham Van Cuong
2: Dr. Nguyen Van Tao
Summary of the new findings of the thesis
First Contribution – Dataset Development:
The dissertation consolidates and extends existing domestic datasets with publicly available data sources to construct a structured, high-quality chest X-ray dataset. The dataset is preprocessed, labeled, and validated by medical experts, standardized in format, resolution, and annotation schema, ensuring its reliability for future machine learning and deep learning research in medical imaging.
Second Contribution – Model Innovation:
The dissertation proposes an improved model for lesion segmentation and classification in occupational silicosis diagnosis based on two integrated approaches:
(1) Adjustment and application of Few-Shot Learning to enable simultaneous segmentation and classification under limited data conditions.
(2) Development of a Graph Transformer Post-hoc (GTP) architecture incorporating Balanced Loss and Ensemble Learning to exploit spatial relationships among feature regions, thereby enhancing model stability and classification performance.
Futher research directions
Building upon the achieved results, several directions for future research are identified:
- Dataset Expansion and Multimodal Integration:
Extending the dataset by incorporating additional data sources, such as CT scans, clinical records, and patient histories, to improve classification accuracy and model robustness. A more diverse dataset will enhance the model’s generalization capability for real-world clinical applications.
- Improving Model Generalization:
Although the current models perform well in controlled settings, they must be further developed to handle heterogeneous and non-standardized data. Investigating multi-task, multi-level, or unsupervised learning strategies may improve robustness and adaptability across different clinical environments.
- Clinical Implementation:
Deploying the proposed models in real-world clinical workflows is both a challenge and an opportunity to validate their efficacy. Future work should involve clinical trials and collaboration with medical practitioners to assess model reliability in assisting the diagnosis of silicosis and potentially other occupational lung diseases.
