Back to 2025 Abstracts
Use of Machine Learning for Ultrasound Assessment of the Ulnar Nerve and Diagnosis of Cubital Tunnel Syndrome
Akhil Dondapati, MD
1; Thomas J Carroll, MD
1; Andrew Rodenhouse, MD
2; Gilbert Smolyak, BS
1; Jeffrey Lillie, PhD
2; Ajay Anand, PhD
1; Lisa Pink, MS
1; Dave J Mitten, MD
1; Constantinos Ketonis, MD, PhD
11University of Rochester Medical Center, Rochester, NY; 2University of Rochester, Rochester, NY
Introduction: Recent studies have established ultrasound, specifically ulnar nerve cross-sectional area (CSA), as a promising tool in the diagnosis of cubital tunnel syndrome (CuTS). The purpose of this study was to develop a machine learning algorithm trained on ultrasound images of the cubital tunnel that can be used to automatically identify and segment the ulnar nerve about the elbow and measure its CSA.
Methods: Healthy control patients were scanned using a Fujifilm-SonoSite ultrasound system, equipped with a high-frequency linear array probe [SLAx with bandwidth 6-13 MHz] by a trained technician or physician. The ulnar nerve was identified and segmented in individual frames from these videos by physicians using a custom graphical user interface application developed in-house. A convolutional neural network (YOLO8) was trained on these images to automatically segment the ulnar nerve, resulting in a binary map (Figure 1). The CSA and dice score was then computed using the binary map of the nerve outline from the prediction and the ground-truth to assess the prediction accuracy.
Results: In total, 34 subjects (11 patients and 23 controls) were imaged with ultrasound, and 2,011 ultrasound grayscale images were segmented from these scans: 1,286 images from 23 subjects (7 patients, 16 controls) were used to train the model, 425 images from 5 subjects (2 patients, 3 controls) were used for validation, and 300 images from 6 subjects (2 patients, 4 controls) for testing. The machine learning model resulted in an average dice score of 0.90 with 296 images (99%) (Figure 2). When comparing the CSA predictions from the machine learning model to the area derived from the ground-truth, the mean and mean absolute difference was 1.78 (13.76%) mm
2 and 2.08 mm
2 (16.02%), respectively.
Conclusions: Ultrasound is an emerging non-invasive modality in the diagnosis of CuTS. We present a novel machine learning algorithm that can accurately identify the ulnar nerve on ultrasound imaging and measure its CSA. Future directions will seek to continue to improve the accuracy of the algorithm and to generate a diagnosis of CuTS through correlation with electrodiagnostic studies, with a goal of implementation in the clinical setting for earlier detection.
Back to 2025 Abstracts
