Enhancing diagnosis: ensemble deep-learning model for fracture detection using X-ray images.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Orthopedic Trauma Analysis

Background:

• Orthopedic trauma frequently causes bone and joint injuries, necessitating accurate radiographic interpretation.
• Radiologists face diagnostic errors due to high volumes of X-ray images, highlighting the need for automated fracture detection.
• Humerus fractures are common orthopedic injuries requiring timely and precise diagnosis.

Purpose of the Study:

• To develop and evaluate an automated ensemble model for detecting humerus bone fractures in X-ray images.
• To improve the accuracy and efficiency of fracture detection in orthopedic radiology.
• To compare the performance of the proposed ensemble model against existing deep learning approaches.

Main Methods:

• An ensemble model integrating MobileNetV2, Vgg16, InceptionV3, and ResNet50 was developed.
• Histogram equalization was used for image preprocessing, and Global Average Pooling for feature extraction.
• The model was trained and validated on the public Mura-v1.1 dataset with an 80:20 split.

Main Results:

• The proposed ensemble model achieved 92.96% accuracy.
• Recall and F1 scores were reported as 91.62% and 92.14%, respectively.
• The model demonstrated superior performance compared to modified deep learning models.

Conclusions:

• The ensemble model effectively automates humerus fracture detection in X-ray images.
• The findings indicate enhanced diagnostic accuracy and efficiency in orthopedic radiology.
• The study benchmarks the ensemble model against state-of-the-art methods, showing significant potential.