Novel transfer learning based bone fracture detection using radiographic images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Machine Learning

Background:

• Bone fractures are common injuries requiring accurate and timely detection.
• Radiographic imaging is standard for fracture assessment, but efficient analysis is critical.
• Early detection of bone fractures is essential for effective treatment and patient outcomes.

Purpose of the Study:

• To develop an efficient neural network method for early bone fracture detection using X-ray images.
• To propose a novel transfer learning approach, MobLG-Net, for enhanced feature engineering in fracture detection.
• To compare the performance of various machine learning models utilizing the novel features generated by MobLG-Net.

Main Methods:

• Spatial features were extracted from bone X-ray images using the MobileNet transfer model.
• Extracted features were processed by a Light Gradient Boosting Machine (LGBM) model to generate class probability features.
• Machine learning models including KNN, LGBM, LR, and RF were trained and evaluated on these novel features with optimized hyperparameters.

Main Results:

• The MobLG-Net approach, combining MobileNet and LGBM, demonstrated superior performance in bone fracture prediction.
• Logistic Regression (LR) and LGBM models trained on MobLG-Net features achieved an accuracy of 99%.
• Cross-validation confirmed the high performance and reliability of the proposed models.

Conclusions:

• The proposed MobLG-Net method significantly improves the accuracy of bone fracture detection from X-ray images.
• This approach offers a promising tool for early and accurate diagnosis, potentially reducing treatment delays and improving patient care.
• The study highlights the effectiveness of transfer learning and gradient boosting for medical image analysis in orthopedics.