基於梅爾倒頻譜係數之非侵入式膝韌帶音訊檢測系統

Non-invasive Detection of Knee Sound Signals for Ligament Diagnosis based on Mel-frequency Cepstrum

蔣亞恬(Ya-Tien Chiang)

指導教授 ： 段裘慶

國立臺北科技大學/電資學院/電腦與通訊研究所/碩士(2014年)

https://doi.org/10.6841/NTUT.2014.00650

摘要

於膝部傷害研究報告中發現韌帶是最常受傷的部位之一，在青少年中受傷的原因屬運動傷害最多，最常見的就是籃球運動導致膝部前十字韌帶斷裂。當肌腱剛斷裂時，若又併發膝部嚴重腫脹，此時醫師單靠觸診有其困難，X光檢查較適用於骨骼的檢查，韌帶等軟組織較不易觀察，而超音波有其侷限性，核磁共振造影則因需長時間等候，則易造成患肢產生肌肉萎縮現象。
為使病患能於門診間讓醫師即時診斷其患肢的病情，本研究以非侵入性檢測方式，利用自製電子聽診器搭配角度提示器進行量測，並以膝關節中之聲音訊號做為判斷，聲音訊號包含關節粗糙度、軟骨破損及其潤滑狀態等可診斷之訊息。本研究藉由對受測者做腿部伸展與屈曲的動作來記錄膝關節聲音訊號，並針對不同韌帶位置設置感測點，再以均方值與濾波等方式，使正常與受損的韌帶聲音訊號差異的幅度更大。接著利用梅爾倒頻譜係數作為膝韌帶聲音訊號之特徵值分解，並進行統計分析再將訊號分類，進而辨識受測者的膝韌帶與半月板是否有損傷。最後以統計學上常使用之精確度、敏感度作為效能分析之評估因子。
本論文所提出之基於梅爾倒頻譜係數之非侵入式膝韌帶音訊檢測系統(KSLDM)系統，與醫院臨床病患診斷資料作互相比較，於半月板部位最佳辨識率以AUC值表示為0.91，而膝韌帶分別於左右腳最佳辨識率則為0.80與0.90。由本研究初步驗證於量測實驗中，取樣23位受測者進行膝韌帶檢測時，具備病徵鑑別度。

關鍵字

膝韌帶聲音訊號 ； 非侵入式 ； 梅爾倒頻譜系數 ； 均方值

並列摘要

The most important joints of human body is knee. Researches about knee injuries, ligaments injuries are always mentioned. Sonography has a limitation for these situations. The MRI examinations are expensive and not always available and time-consuming. In this study, we conducted a non-invasive method to detect ligament wheather it is fracture or laceration. The detection tool is an electronic stethoscope made by myself. We detect and record the knee-joint sound to determine the symptom of knee. The knee-joint sound includes some diagnostic messages related to the roughness, breakdown and lubrication status. The subjects’ knee-joint sound was recorded during the motion of extension and flexion of leg. Differing from other methods, we increase the positions of sensing on the knee point. Then we applied band pass filters to keep the required frequency range. Using the mean square method could make the sound signals of stricken ligament more distinct than normal sound signals. Adopting the Mel-frequency Coefficients as the characteristic value of sound, thus classify sound signals to judge whether the subject’s ligament get hurt and the extent of damage. We conducted experiments and evaluated three factors: sensitivity, precision and ROC curve. The results suggested that the proposed KSLDM had AUC higher than 0.91 in meniscus identification and had 0.8,0.9 of left and right ligament identification. In this study, the KSLDM showed good recognition performance under 23 subjects being detected with the knee sound signal.

並列關鍵字

Knee sound signal ； Non-invasive ； Mel-frequency ； Mean-square

參考文獻

[1] 康翠籃，"前十字韌帶重整手術後的復健"，大專體育，vol. 35，1998，pp. 127-133。

[2] 張世緯、張怡雯，"前十字韌帶損傷對下肢運動生物力學的影響"，大專體育，vol. 112，2011，pp. 69-75。

[3] 曾薇、王桂芸，"前十字韌帶重建手術對病人之身心影響及其照護"，榮總護理，vol. 24，no. 3，2007，pp. 244-250。

[4] J. Ahovuo, P. Paavolainen, and P. Slatis, "Diagnostic Value of Sonography in Lesions of the Biceps Tendon," Clinical Orthopaedics and Related Research, vol. 202, 1986, pp. 184-188.

[5] B. P. Boden, G. S. Dean, J. A. Feagin, and W. E. Garrett, "Mechanisms of Anterior Cruciate Ligament Injury" Thorofare Orthopedics, vol. 23, no. 6, 2000, pp. 573-578.