Performance Analysis of Feature-Based Transfer Learning Using VGG19 for Detecting Hairline and Spiral Fractures

Correct classification of bone fractures is crucial for accurate medical treatment, as traditional diagnostic procedures can be time-consuming and prone to errors. This study explores the use of the VGG19 architecture combined with three classifiers, namely Support Vector Machine (SVM), Logistic Regression (LR), and k-Nearest Neighbors (kNN) for classifying hairline and spiral fractures in X-ray images. A dataset of 400 X-ray images was utilized, from which features were extracted using the VGG19 model. These features were then used to train the SVM, LR, and kNN classifiers. Among the models tested, the VGG19-LR pipeline demonstrated the best overall performance, achieving high accuracy and robustness in both validation and testing phases. The VGG19-SVM model also showed strong performance but was slightly less effective than the VGG19-LR. In contrast, the VGG19-kNN model yielded the weakest results, indicating lower suitability for this classification task. These results suggest that the VGG19-LR pipeline was suitable for the given dataset. The high accuracy demonstrated indicates that transfer learning can offer an efficient method for classifying bone fractures, particularly for hairline and spiral fracture, in contrast to training a deep neural network from the beginning. This enables the creation of automated bone fracture diagnosis using computer vision and deep learning.