Predicting Surgical Versus Non-Surgical Management of Acute Isolated Distal Radius Fractures in Patients Under Age 60 Using a Convolutional Neural Network

Distal radius fractures (DRF) represent up to 20% of fractures in the emergency department. Delays to surgery greater than 14 days are associated with poorer functional outcomes and increased healthcare utilization/costs. At our institution, the average time to surgery is greater than 19 days due to separation of surgical and nonsurgical care pathways and a lengthy referral process. To address this challenge, we aimed to create a convolutional neural network (CNN) capable of automating DRF x-ray analysis and triaging. We hypothesize this model will accurately predict whether an acute isolated DRF fracture in a patient under age 60 will be treated surgically or nonsurgically based on radiographic input. We included 163 patients (93 surgical, 70 nonsurgical) under age 60 who presented to the ED between 2018–2023 with an acute isolated DRF and were referred for clinical follow up. Radiographs taken within 4 weeks of their injury were collected in posterior-anterior and lateral views before being pre-processed for model training. The gold standard for diagnosis equivalence was the surgeon's clinical decision of treating the patient surgically or nonsurgically. 723 radiographic pairs were used for model training. The best-performing model was obtained using 7 CNN layers, 1 fully connected (FC) layer, an image input size of 256x256 pixels, and a 1.5x weighting for volarly displaced fractures. Model performance metrics at the per-scan and per-patient level were evaluated (Table 1). Values for True Positive, True Negative, False Positive and False Negative were calculated (Figure 3). A CNN-based algorithm can predict with 88% accuracy whether treatment of an acute isolated DRF in a patient under age 60 will be treated surgically or nonsurgically. By promptly identifying patients who would benefit from expedited surgical treatment pathways, this model can reduce wait times and subsequently enhance patients' outcomes. For any figures or tables, please contact the authors directly.