Abstract:
Bone fractures of the arm are difficult to diagnose with precision and speed in the field
of radiology, which greatly impacts the healthcare system considering the available
treatment options. This problem was approached by building an AI system called
Boneguard, an arm fracture detection system that autonomously detects and classifies
fractures through radiographic images. We utilized the MURA dataset containing
multiple musculoskeletal X-ray images to train and evaluate our models for precision
and diagnosis dependability with the set aim of achieving accuracy. To ensure the
models would perform on different classes of images, we designed a pre-processing
pipeline that incorporated image resizing, normalization, and advanced data
augmentation techniques such as rotation,. flipping, and zooming. These steps
improved the uniformity of the dataset and allowed the models to generalize to
variations in real-life. According to the findings that were achieved during the test
phase, Boneguard could improve the accuracy and efficiency of radiology workflows
to the benefit of clinicians and patients. To make the platform scalable and reliable in
actual clinical situations, we need to continue testing the models with fresh data, add
explain ability, and enhance the models to test them.
