Background Road Traffic Crash Injuries contribute to significant morbidity and mortality worldwide, specifically in LMIC’s. Artificial Intelligence (AI), and Machine Learning (ML) has a vast potential to exponentially optimize health care research. The utility of AI and ML in effectively boosting the trauma care systems has yet to be explored. Very limited studies report the utility of ML and AI in predicting Injuries through utilizing multiple XRays(radiology). X-ray Trauma Series(C-spine, Chest and Pelvic Xrays) are a modality used in Clinical ED practice for diagnosing injuries in trauma victims.
Aims To develop a Machine learning tool that can predict and detect injuries utilizing X-Ray trauma series(Tri-Image modality) effectively and efficiently through image recognition technique using CNN deep learning approach.
Methods This cross sectional pilot study was conducted at the Aga Khan University Hospital, Karachi. The data for this study was extracted retrospectively from medical records of Road Traffic Crash Victims. We applied CNN computer vision pipeline to an annotated image dataset of 150 normal and 30 abnormal images. Image annotation was performed using Computer Vision annotation tool(CVAT).The images were then augmented and preprocessed. The final model was trained using Roboflow Train(Auto-ML ) tool customized by using CNN with tensorflow at the backend.
Results Our model was able to classify injuries with 91.9%of validation set accuracy suing CNN computer vision classification technique.
Conclusion Our pilot results were a useful test of hypothesis of tri-modal injury image data prediction. We believe that with this approach if applied to a larger data-set can be a useful prediction tool of injuries and can be implemented on light weight devices for use in low resource settings
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.