X-Ray Imaging Dataset for Detecting Fractured vs. Non-Fractured Bones

This study uses a curated dataset of human bone X-ray images collected from two hospital laboratories, consisting of two classes: fractured and non-fractured bones. The dataset contains 420 original images (130 fractured and 290 non-fractured), with variations in size and resolution. All samples were carefully labeled and clinically verified by Dr. Mohammad Saiful Malek, MBBS, BCS (Health), Consultant and Surgeon at the UHC, Fulbaria Government Health Hospital, Mymensingh, Bangladesh, ensuring reliable annotations for research.

To improve model robustness and generalization, data augmentation was applied only due to the limited number of samples. Using the Albumentations library, images were resized to 512×512 pixels and transformed using horizontal and vertical flips, brightness–contrast adjustments, rotations, and affine shearing. Each original sample produced six augmented versions, expanding the dataset from 420 images to 2520 images (780 fractured and 1740 non-fractured). After augmentation, combining spatial enhancement methods were additionally applied to further improve structural clarity and feature representation. This combined strategy increased data variability, balanced the dataset, and significantly enhanced the performance of the proposed model.

The dataset is organized into the following class folders:
- Fractured
- Non-Fractured

Original images:
- Fractured (130)
- Non-Fractured (290)