Year: 2025
Type: conference-paper
Abstract: Pelvic fractures, which are critically important due to their impact on vital structures, require precise detection for effective treatment. The research uses pre-training to accurately identify pelvic fractures in X-ray images. In this study, pelvic fractures are discussed as a key area in orthopedics related to object detection using AI in medical imaging. The methods involve collecting nearly 1000 pelvic X-ray images from various hospitals and preprocessing them with augmentation techniques. The pre-trained models, including a custom CNN, InceptionV3, ResNet50, and ResNet101, leverage their deep architectures to improve feature extraction. These models were trained and evaluated using metrics such as accuracy, precision, recall, and F1-score. Results show that ResNet101 achieves the highest testing accuracy (0.8000) and precision (0.8179), despite longer training times (1817.57s). The custom CNN trains faster (762.40s) but attains lower accuracy (0.6643). Future research will investigate ensemble learning, attention mechanisms, and larger datasets to enhance generalizability and facilitate real-time clinical deployment.
Language: English
Topic: Medical Imaging and Analysis
Subfield: Biomedical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W7204477239