應用震動感測儀(G-sensor)與人工智慧(深度學習神經網路)

協助診斷排尿障礙

蔡芳生¹、彭元宏¹²、鄭子維³、蔡鈺鼎⁴

敏盛綜合醫院 泌尿科；¹國立台灣大學附設醫院 泌尿部；²天成醫院 泌尿科；³逢甲大學 機械航太工程博士學程；⁴逢甲大學 電聲碩士學位學程

Application of Artificial Intelligence (Deep Learning Neural Network) for Assistant Diagnosis of Voiding Dysfunction by Using a Vibration Sensor

Vincent F.S. Tsai¹, Yuan-Hung Pong^1,2, Zi-Wei Zheng³, Yu-Ting Tsai⁴

Divisions of Urology, Min-Sheng General Hospital, Taoyuan, Taiwan; ¹Department of Urology, National Taiwan University Hospital, Taipei, Taiwan; ²Divisions of Urology, Ten-Chen General Hospital, Taoyuan, Taiwan; ³Ph.D. Program of Mechanical and Aeronautical Engineering, Feng Chia University, Taichung, Taiwan; ⁴Master’s Program of Electro-Acoustics, Feng Chia University, Taichung, Taiwan

Purpose: With increasing number of people having voiding dysfunction, the voiding pattern of patients is measured mainly and only by voiding diary at home and urodynamic examinations in hospitals. Voiding diary is less objective and often contains missed data, while the latter lacks frequent measurements and is a relatively invasive procedure. For these unmet needs, this study developed an innovative and contact-free technology that provided voiding pattern measurement and assistant diagnosis.

Materials and Methods: Vibration signals during urination were first detected using an accelerometer (G-sensor) and then converted into the mel-frequency cepstrum coefficient (MFCC). Lastly, an artificial intelligence model combined with uniform manifold approximation and projection (UMAP) dimensionality reduction was used to analyze and predict six common patterns of uroflowmetry to assist in diagnosing voiding dysfunction. The model was applied to the voiding database, which included data from 76 males aged 30 to 80, who also received uroflowmetry for voiding symptoms at the same time.

Results: The resulting system accuracy (precision, recall, and f1-score) was around 98% for both the weighted average and macro average. There are associations between G-sensor signal and void-volume, void-time, maximal flowrate (p<.001).