TY - JOUR
T1 - AI for Automated Segmentation and Characterization of Median Nerve Volume
AU - Jagtap, Jaidip M.
AU - Kuroiwa, Tomoyuki
AU - Starlinger, Julia
AU - Farid, Mohammad Hosseini
AU - Lui, Hayman
AU - Akkus, Zeynettin
AU - Erickson, Bradley J.
AU - Amadio, Peter
N1 - Publisher Copyright:
© 2023, Taiwanese Society of Biomedical Engineering.
PY - 2023/8
Y1 - 2023/8
N2 - Purpose: Carpal tunnel syndrome (CTS) is characterized anatomically by enlargement of the median nerve (MN) at the wrist. To better understand the 3D morphology and volume of the enlargement, we studied its volume using automated segmentation of ultrasound (US) images in 10 volunteers and 4 patients diagnosed with CTS. Method: US images were acquired axially for a 4 cm MN segment from the proximal carpal tunnel region to mid-forearm in 10 volunteers and 4 patients with CTS, yielding over 18,000 images. We used U-Net with ConvNet blocks to create a model of MN segmentation for CTS study, compared to manual measurements by two readers. Results: The average Dice Similarity Coefficient (DSC) on the internal and external validation datasets was 0.82 and 0.81, respectively, and the area under the curve (AUC) was 0.92 and 0.88, respectively. The inter-reader correlation DSC was 0.83, and the AUC was 0.98. The correlation between U-Net and manual tracing was best when the MN was near the surface. A US phantom mimicking the MN, imaged at varied scanning speeds from 7 to 45 mm/s, showed the volume measurements were consistent. Conclusion: Our AI model effectively segmented the MN to calculate MN volume, which can now be studied as a potential biomarker for CTS, along with the already established biomarker, cross-sectional area.
AB - Purpose: Carpal tunnel syndrome (CTS) is characterized anatomically by enlargement of the median nerve (MN) at the wrist. To better understand the 3D morphology and volume of the enlargement, we studied its volume using automated segmentation of ultrasound (US) images in 10 volunteers and 4 patients diagnosed with CTS. Method: US images were acquired axially for a 4 cm MN segment from the proximal carpal tunnel region to mid-forearm in 10 volunteers and 4 patients with CTS, yielding over 18,000 images. We used U-Net with ConvNet blocks to create a model of MN segmentation for CTS study, compared to manual measurements by two readers. Results: The average Dice Similarity Coefficient (DSC) on the internal and external validation datasets was 0.82 and 0.81, respectively, and the area under the curve (AUC) was 0.92 and 0.88, respectively. The inter-reader correlation DSC was 0.83, and the AUC was 0.98. The correlation between U-Net and manual tracing was best when the MN was near the surface. A US phantom mimicking the MN, imaged at varied scanning speeds from 7 to 45 mm/s, showed the volume measurements were consistent. Conclusion: Our AI model effectively segmented the MN to calculate MN volume, which can now be studied as a potential biomarker for CTS, along with the already established biomarker, cross-sectional area.
KW - Carpal tunnel syndrome
KW - Cross-sectional area
KW - Machine learning
KW - Median nerve
KW - U-Net
KW - Ultrasound
UR - http://www.scopus.com/inward/record.url?scp=85166621410&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85166621410&partnerID=8YFLogxK
U2 - 10.1007/s40846-023-00805-z
DO - 10.1007/s40846-023-00805-z
M3 - Article
AN - SCOPUS:85166621410
SN - 1609-0985
VL - 43
SP - 405
EP - 416
JO - Journal of Medical and Biological Engineering
JF - Journal of Medical and Biological Engineering
IS - 4
ER -