BDCXR-3257: 3,257 Chest X-Rays with Pneumonia Labels from Bangladesh
by Sabbir Hossain Durjoy·Updated 1mo ago
880.2 MB1files
Available on 1 platform
Sign in to view source links and access this dataset
Description
3,257 clinically verified chest X-ray images collected from Dr. M.R. Khan Shishu Hospital & Institute of Child Health in Dhaka, Bangladesh, between 2024 and 2025. The dataset was created by Sabbir Hossain Durjoy and contains images categorized into Normal, Bacterial Pneumonia, and Viral Pneumonia classes. All images were independently annotated by two expert clinicians and underwent a preprocessing pipeline for quality and consistency.
Use Cases
Training binary classifiers for pneumonia detection based on chest X-ray images.
Developing multi-class models to differentiate between bacterial and viral pneumonia etiologies.
Benchmarking model performance on a dataset with expert clinician annotations and high inter-annotator agreement.
Studying the generalizability of diagnostic AI models on a demographically diverse, paediatric South Asian population.
Strengths
Contains 3,257 chest X-ray images, providing a substantial volume for model training.
Features expert clinician annotations with a reported Fleiss’ κ > 0.95, indicating near-perfect inter-annotator agreement.
Includes a preprocessing pipeline (text removal, BPDFHE, manual cropping) applied for image quality and consistency.
Explicitly focuses on a paediatric population from a South Asian clinical setting, addressing a noted gap in existing public datasets.
Limitations
Row count for the primary data is unknown, which may limit suitability assessment for specific modeling tasks.
Column-level documentation is absent; field semantics and image metadata must be inferred after download.
The dataset's geographic scope is limited to a single hospital in Dhaka, Bangladesh, which may affect model generalizability to other regions.
Provenance
Source
Dr. M.R. Khan Shishu Hospital & Institute of Child Health, Dhaka, Bangladesh.
Collection Method
Chest X-ray images were selected from a broader collection of medical images, annotated independently by two expert doctors, and preprocessed.
Time Range
2024 to 2025
Freshness
Last updated 2026-04-30 20:03:11.
Geography
Dhaka, Bangladesh
Dataset is released under a CC-BY-4.0 license. The primary data is packaged in a ZIP file (880.2 MB).