居家復健用之穿戴式系統的設計與實作

蕭仲琦

《TANET2017 臺灣網際網路研討會》 2017年 (2017/10) Pp. 482-488

https://doi.org/10.6728/TANET.201710.0084

摘要

復健治療的過程經常較長且病患需忍受過程中的疼痛，導致病患按時到醫院作復健的比例不高。本論文提出一套可以連網路的居家肩關節復健系統，針對五十肩病患，讓病患能早日康復。本系統包含自行研製的硬體與軟體，硬體部分採用Android智慧型手機搭配另一個以Arduino製作的穿戴式裝置分別佩掛於手臂關節處，將兩種裝置的感測資料彙整作前處理。前處理程序確保資料校準並正規化，將資料輸入手機APP內部實作之「倒傳遞類神經網路」運算、判斷復健動作正確與否，隨即再將結果傳送到後端資料庫加以儲存。醫療人員可以透過本論文開發網頁以圖形化的方式呈現復健動作之正確率，降低醫療人員的負擔。本系統目前考慮「鐘擺運動」和「抬手運動」二種復健動作，並以三位體型不同的受測者就每一種動作分別以正確與錯誤的復健行為各操作20次以進行實測，實驗結果顯示本系統可精準地辨識鐘擺運動達91.67%、辨識抬手運動達100%。

關鍵字

Android ； 倒傳遞類神經網路 ； Bluetooth ； Arduino ； MySQL

並列摘要

Rehabilitation is typically a long process, during which patients often need to endure pains and sufferings. Such hardship makes people inconvenient to undergo rehabilitation, so only a small population of patients go to hospitals for regular treatment. In view that 87% of frozen shoulder (adhesive capsulitis of the shoulder) can recover completely through regular rehabilitation, this thesis aims at the promotion of shoulder rehabilitation. To this end, we propose a rehabilitation system for residential use that can free the user of the burden of repeatedly going to medical institutions back and fro, so as to facilitate early recovery. Our developed system consists of hardware and software. The hardware component involves an Android smartphone and a self-developed wearable device over Arduino that are attached close to the wrist and elbow joints respectively. Both devices collect in synergy the data of arm physical movements while rehabilitation is in progress. The Android APP combines its own sensed data with the receipts for subsequent preprocessing of data alignment and normalization. Preprocessed outputs are then fed to another procedure implementing a back-propagation neural network which has been trained to resolve if the fed data corresponds to a correct motion. Next the APP transmits the resolved outcomes over the Internet to a backend database so that medical personnel are enabled to keep track of statistical results illustrating how individuals carried out rehabilitation. Currently we consider two types of exercise: pendulum and arm-raising exercises. Experiments were conducted by letting three subjects of different habitus, for each exercise, practice 10 correct and 10 incorrect rehabilitation movements respectively. Experimental results indicate that our development can identify the pendulum exercise at success rate of 91.67% and the arm raising exercise at 100%. The misjudgment rate is comparatively low, reflecting the value and usefulness of our design and implementation.

並列關鍵字

Android ； Back Propagation Neural Network ； Bluetooth ； Arduino ； MySQL