Related Experiment Video
Updated: Jun 11, 2026
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
Development and Preliminary Evaluation of an EfficientNet-Based Deep Learning System for Ultrasound Assessment of
Wei Ding Wang1, Siew-Ying Mok1, Yang Mooi Lim2
1Department of Mechatronics and Biomedical Engineering, Lee Kong Chian Faculty of Engineering and Science, Universiti Tunku Abdul Rahman, Bandar Sungai Long, Kajang 43000, Selangor, Malaysia.
This study developed a machine learning model using ultrasound images to accurately diagnose neck disorders, showing high performance in assessing lower cervical regions and fascial abnormalities. This AI-driven approach offers efficient and consistent clinical decision support for neck pain assessment.
Area of Science:
- Medical imaging and diagnostics
- Artificial intelligence in healthcare
- Musculoskeletal disorder assessment
Background:
- Neck disorders significantly impact quality of life, with traditional diagnostics relying on variable clinician expertise.
- Current diagnostic methods for neck disorders, including ultrasound, have limitations, especially in assessing fascial abnormalities.
- Machine learning applications for diagnosing neck disorders via ultrasound are underexplored.
Purpose of the Study:
- To develop and assess a machine learning model for diagnosing neck disorders using ultrasound images.
- To address challenges of limited labeled ultrasound data for neck disorders.
- To provide accurate and efficient diagnostic support for neck disorders.
Main Methods:
- Utilized ultrasound images from 184 patients to train machine learning algorithms.
- Employed EfficientNet with transfer learning to overcome data scarcity and enhance model generalizability.
- Implemented 5-fold cross-validation with class weighting and AdamW optimizer for robust model training.
Main Results:
- The machine learning model achieved high performance in classifying neck disorders.
- Highest weighted average F1-scores of 76% and 81% were recorded for deep fascia fuzzy texture and deep fascia/myofascial adhesion in lower cervical regions.
- Macro averages indicated consistent class-wise accuracy, with scores of 74% and 78% respectively.
Conclusions:
- The developed model demonstrates robust classification performance for neck disorder assessment, particularly in the lower cervical region.
- This AI-based approach has the potential to enhance clinical decision-making through consistent and accurate diagnostic assistance.
- Further validation in diverse clinical settings is recommended to improve real-world applicability.
