Automated fracture screening using an object detection algorithm on whole-body trauma computed tomography.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in radiology
• Trauma care diagnostics

Background:

• Emergency departments face diagnostic errors in trauma care, especially for fractures.
• Deep learning, specifically Convolutional Neural Networks (CNNs), shows promise in improving medical diagnostic accuracy and efficiency.
• The application of CNNs for fracture detection in whole-body CT scans requires further investigation.

Purpose of the Study:

• To evaluate the efficacy of CNNs for automatic localization and classification of pelvic, rib, and spine fractures.
• To determine if a CNN-based fracture detection algorithm can assist physicians in diagnosing fractures.
• To assess the impact of CNN assistance on diagnostic sensitivity and interpretation time for orthopedic surgeons.

Main Methods:

• Utilized 7664 whole-body CT axial slices from 200 trauma patients.
• Developed and applied a CNN deep learning model for automatic fracture detection.
• Calculated sensitivity, precision, and F1-score to evaluate the CNN model's performance.

Main Results:

• The CNN model achieved a sensitivity of 0.786, precision of 0.648, and F1-score of 0.711 for grouped pelvic, spine, and rib fractures.
• Surgeons demonstrated improved sensitivity in fracture detection when assisted by the CNN model.
• Significant reduction in CT scan reading and interpretation time was observed, particularly for less experienced surgeons.

Conclusions:

• CNN-based fracture detection is applicable to whole-body CT images for identifying pelvic, rib, and spine fractures.
• Assistance from CNN models can reduce missed fractures and expedite diagnosis in polytrauma patients.
• The integration of CNNs into clinical workflows has the potential to enhance patient care through more efficient and accurate fracture diagnosis.