MilitaryHospital175/VNMedical_bv175

ViX-Ray: A Vietnamese Chest X-Ray Dataset for Vision-Language Models

ViX-Ray is the first publicly available Vietnamese chest X-ray dataset designed for vision-language model (VLM) research in medical AI. It pairs radiographic images from Vietnamese patients with expert-written clinical annotations in Vietnamese, directly addressing the lack of Vietnamese medical data in existing VLMs.

• License: CC BY-NC-SA 4.0
• Paper: arXiv:2603.15513

---

Dataset Overview

Each sample contains a PA-view chest X-ray image paired with Vietnamese-language radiology reports written by certified radiologists. All Protected Health Information (PHI) has been removed; clinically relevant metadata (age, gender) is retained. The dataset was collected under Institutional Review Board (IRB) approval.

Splits

Split	Samples	Percentage
Train	4,320	80%
Val	540	10%
Test	540	10%
Total	5,400	100%

Fields

Field	Description
image	PA-view chest X-ray image
findings	Detailed descriptive narrative of anatomical/pathological observations (Vietnamese)
impressions	Clinical diagnostic conclusions (Vietnamese)
Ages	Patient age in years
sex	Patient gender
view	Imaging perspective (PA — Posterior Anterior)
Ghi chú	Procedure description (Vietnamese)
Kỹ thuật	Technical details of the radiograph

Report Text Length (tokens)

Split	Findings (avg)	Impressions (avg)
Train	46	12
Val	45	14
Test	46	13

Average patient age: ~69–70 years across all splits.

---

Common Findings

Most frequently mentioned anatomical structures: Heart (tim), Lungs (phổi), Ribs (xương sườn), Diaphragm (vòm hoành), Pulmonary hilum (rốn phổi)

Most common diagnoses: Interstitial lung disease, Cardiomegaly, Pulmonary infiltrates/consolidations, Rib fractures, Aortic calcifications

---

Intended Use

• Training and fine-tuning vision-language models for Vietnamese medical image analysis
• Benchmarking multilingual and Vietnamese-specific VLMs
• Research on hallucination reduction in clinical AI
• Multi-turn clinical diagnostic conversation modeling

---

Benchmark Results

Five open-source VLMs were fine-tuned and evaluated on three stages: findings generation → impression generation → multi-turn clinical dialogue. Best performing model: Qwen2.5-VL-7B (ROUGE-1: 84.30% on findings, 95.20% on multi-turn impressions).

---

Citation

If you use this dataset, please cite:

@article{nguyen2026vix,
  title={ViX-Ray: A Vietnamese Chest X-Ray Dataset for Vision-Language Models},
  author={Nguyen, Duy Vu Minh and Truong, Chinh Thanh and Tran, Phuc Hoang and Le, Hung Tuan and Dat, Nguyen Van-Thanh and Pham, Trung Hieu and Van Nguyen, Kiet},
  journal={arXiv preprint arXiv:2603.15513},
  year={2026}
}