DeepFracture: A Fine Tuned Transfer Learning based Model for Bone Fracture Classification

This work introduces DeepFracture, a transfer learning aided model which is used to classify ten different kinds of bone fractures using pre-trained architecture models VGG16, MobileNetV2, and InceptionV3. Downloaded from Kaggle, the dataset was resized and augmented as preprocessing to mitigate the issue of class imbalance. Techniques of Fine-tuning the model with respect to its architecture included: freezing certain layers of the model and learning rate up to 0.001 was reduced to 0.0001. At the end of the training MobileNetV2 was able to give the highest accuracy of 97.40%. These results prove that transfer learning can be used effectively in classifying medical images as it guarantees accurate fracture detection in a short time. This is beneficial to the healthcare practitioners as it minimizes the risks of misdiagnosis and enhances the effectiveness of treatment. Future studies might be centered on increasing the data set and applying these models in the clinical settings in real time to assist in making medical decisions.