Abstract:
Pubic rami fractures are common in the elderly, often resulting from low-energy trauma, and can
be very challenging to diagnose because they are subtle on X-rays. It is crucial to accurately
identify this condition at the right time to prevent complications such as immobilisation and
chronic pain. The proposed study involves a pretrained convolutional neural network (CNN)
based automated detection system for pubic rami fractures in X-ray images, which will be collected
from public hospitals in Sri Lanka. The dataset, consisting of 2,000 X-ray images labelled
as fractured or non-fractured, was preprocessed and augmented to improve the model’s robustness.
Four pretrained CNN models, such as ResNet-50, ResNet-101, EfficientNet-B0, and EfficientNetV2,
were fine-tuned for binary classification of pubic rami fractures. The ResNet-101
model showed the best performance, achieving an accuracy of 82%, precision of 0.84, recall
of 0.82, and F1 score of 0.83, outperforming other models and even the custom CNN baseline.
These results highlight the potential of pretrained CNNs to improve detection accuracy through
transfer learning, especially given the scarcity of medical image data. The system could assist
clinicians in early diagnosis, reduce diagnostic errors, and streamline clinical workflows.
However, challenges such as unbalanced class representation and the subtlety of fracture signs
remain, indicating that larger and more diverse datasets, along with more advanced augmentation
methods, will be necessary in future studies.
