Predicting Surgical Versus Nonsurgical Management of Acute Isolated Distal Radius Fractures in Patients Under Age 60...

Area of Science:

• Orthopedic Surgery
• Artificial Intelligence in Medicine
• Radiology

Background:

• Distal radius fractures (DRFs) are common emergency department visits.
• Delays in surgical treatment (over 14 days) correlate with worse patient outcomes and increased healthcare costs.
• Current referral processes at our institution lead to average surgical delays exceeding 19 days.

Purpose of the Study:

• To develop a convolutional neural network (CNN) for automated analysis and triage of DRF X-rays.
• To predict whether acute isolated DRFs in patients under 60 will require surgical or non-surgical treatment based on radiographic data.

Main Methods:

• Utilized 723 radiographic image pairs (PA and lateral views) from 163 patients under 60 with acute isolated DRFs (2018-2023).
• Trained a CNN model (seven CNN layers, one fully connected layer, 256x256 input size) with a 1.5x weighting for volarly displaced fractures.
• Surgeons' treatment decisions served as the reference standard for model accuracy assessment.

Main Results:

• The optimized CNN model achieved 88% overall accuracy and 100% sensitivity in predicting surgical vs. non-surgical treatment.
• Specific performance metrics included 100% true positive, 72.7% true negative, 27.3% false positive, and 0% false negative rates.
• The model demonstrated high efficacy in classifying fracture treatment needs.

Conclusions:

• A CNN-based algorithm, trained on institution-specific data, can accurately predict surgical or non-surgical treatment for DRFs in patients under 60.
• This AI tool has the potential to expedite surgical referrals by identifying suitable candidates, thereby reducing treatment delays and improving patient care pathways.