Quantitative Evaluation of Tendon Gliding Sounds and Their Classification Using Deep Learning Models

Abstract

This study aims to develop and evaluate a deep learning (DL) model for classifying tendon gliding sounds recorded using digital stethoscopes (Nexteto, ShareMedical, Japan, Nagoya). Specifically, we investigate whether differences in tendon excursion and biomechanics produce distinct acoustic signatures that can be identified through spectrogram analysis and machine learning (ML). Tendon disorders often present characteristic tactile and acoustic features, such as clicking or resistance during movement. In recent years, artificial intelligence (AI) and ML have achieved significant success in medical diagnostics, particularly through pattern recognition in medical imaging. Leveraging these advancements, we recorded tendon gliding sounds from the thumb and index finger in healthy volunteers and transformed these recordings into spectrograms for analysis. Although the sample size was small, we performed classification based on the frequency characteristics of the spectrograms using DL models, achieving high classification accuracy. These findings indicate that AI-based models can accurately distinguish between different tendon sounds and strongly suggest their potential as a non-invasive diagnostic tool for musculoskeletal disorders. This approach could offer a non-invasive diagnostic tool for detecting tendon disorders such as tenosynovitis or carpal tunnel syndrome, potentially aiding early diagnosis and treatment planning.

Keywords: artificial intelligence; deep learning; digital stethoscopes; machine learning; tendon gliding sounds.

Figure 1. Location of tendon sound recording using a digital stethoscope.

Figure 2. Tendon gliding sound recordings of the thumb using Audacity.

Audacity: Audacity Team, Pittsburgh, USA

Figure 3. Tendon gliding sound recordings of the index finger using Audacity.

Audacity: Audacity Team, Pittsburgh, USA

Figure 4. Spectrogram of tendon gliding sounds for the thumb.

Figure 5. Spectrogram of tendon gliding sounds for the index finger.

Figure 6. The dataset displays spectrograms of tendon gliding sounds collected from the thumb and index finger during flexion-extension movements.

Figure 7. The confusion matrices (a: MobileNet, b: ResNet101, c: EfficientNet).

A confusion matrix is a performance evaluation tool that visualizes the classification results by comparing predicted labels with actual labels. It indicates the number of correctly and incorrectly classified cases for each class, allowing detailed assessment of model performance beyond overall accuracy.

Figure 8. The receiver operating characteristic (ROC) curves (a: MobileNet, b: ResNet101, and c: EfficienNet).

The receiver operating characteristic (ROC) curve illustrates the relationship between the true positive rate and false positive rate, providing a visual evaluation of the model’s classification performance. An AUC (area under the curve) value closer to 1 indicates higher classification accuracy.

Figure 9. Comparison of original spectrogram and occlusion sensitivity analysis results.

The figure shows the original spectrograms of sound data recorded during flexion and extension movements, along with the results of occlusion sensitivity analysis performed by MobileNet_v2, ResNet101, and EfficientNet. The thumb is labeled as A, and the index finger as B. The occlusion sensitivity analysis confirmed that each model focused on the high-frequency bands during thumb flexion and extension movements.