From Pixels to Diagnosis: A Systematic Review of Deep Learning in Femoral Fracture Detection and Classification

Femoral fractures are becoming more common and require fast and accurate diagnosis, which makes them a significant worldwide health concern for older people. It is a major global health problem because it is becoming more common and needs a quick and accurate diagnosis. Traditional X-ray image interpretation risks human error and a lack of consistency, mainly in emergencies. To address these challenges, this review paper explores the development and application of deep learning (DL) techniques, using convolutional neural networks (CNNs) and Vision Transformers (ViTs), for automated femur fracture detection and classification using X-ray and CT imaging. Several models showed excellent diagnostic performance: the Faster R-CNN achieved a multi-class accuracy of 90% with an IoU of 0.87, the ViTs achieved an accuracy of 92% with an AUC of 0.94, and the ResNet50 achieved up to 95% accuracy. Advanced techniques like curriculum learning, attention mechanisms, and data augmentation with GANs have further enhanced the robustness and interpretability of the model. Although these approaches can help radiologists to accurately and quickly recognize fractures, there are limitations in dataset uniformity, transparency, and real-world integration. Clinical adoption requires further study.