Bone Fracture Detection Model Data

About

Model Purpose: Efficient and accurate identification of bone fractures using deep learning techniques, primarily centered on X-ray images.

Data Source: MURA dataset, consisting of 20,335 radiographs of the musculoskeletal system, categorized based on three distinct bone parts: the elbow, hand, and shoulder.

Features Used: X-ray images of the musculoskeletal system. Preprocessing includes operations like horizontal flipping and data augmentation for enhanced diversity.

Model Architecture: The algorithm utilizes a ResNet50 neural network to classify the bone type depicted in the image. Three distinct models are employed to recognize fractures in specific types of bones.

Dataset Overview

PartNormalFracturedTotal
Elbow316022365396
Hand433016736003
Shoulder449644408936

Model Performance

  • Elbow Training accuracy - 89%
  • Hand Test accuracy - 83.15%
  • Shoulder Validation accuracy - 82.56%
  • Elbow Test accuracy - 79.33%
  • Hand Validation accuracy - 76.3%
  • Shoulder Test accuracy - 74.2%