A Convenient Non-harm Cervical Spondylosis Intelligent Identity method based on Machine Learning

Cervical spondylosis (CS), a most common orthopedic diseases, is mainly identified by the doctor’s judgment from the clinical symptoms and cervical change provided by expensive instruments in hospital. Owing to the development of the surface electromyography (sEMG) technique and artificial intelligence, we proposed a convenient non-harm CS intelligent identify method EasiCNCSII, including the sEMG data acquisition and the CS identification. Faced with the limit testable muscles, the data acquisition method are proposed to conveniently and effectively collect data based on the tendons theory and CS etiology. Faced with high-dimension and the weak availability of the data, the 3-tier model EasiAI is developed to intelligently identify CS. The common features and new features are extracted from raw sEMG data in first tier. The EasiRF is proposed in second tier to further reduce the data dimension, improving the performance. A classification model based on gradient boosted regression tree is developed in third tier to identify CS. Compared with 4 common machine learning classification models, the EasiCNCSII achieves best performance of 91.02% in mean accuracy, 97.14% in mean sensitivity, 81.43% in mean specificity, 0.95 in mean AUC.

So we select 7 representative movements from the functional activities above which consist of the following movements: bow(A 1 ), head backwards(A 2 ), left flexion(A 3 ), right flexion(A 4 ), left rotation(A 5 ), right rotation(A 6 ), hands up(A 7 ).
The instruction of data acquisition We use the 6-channel sEMG device as shown in Figure S 2 , each channel of which is connected to fixed muscle. The 6-channel sEMG device can simultaneously collect 6 sEMG signals from 6 muscles with a sampling frequency of up to 1062 Hz.
We acquire the data according to the following instruction. Firstly, we connect the sEMG device to laptop. Secondly, the selected muscles are connected with sEMG device by surface electrodes according to the Figure S  • Standing: open your feet so that the distance between your feet is shoulder width. Keep head up, eyes looking straight ahead, and arms naturally hanging by sides.
• A 1 : stand for 5 seconds, bow to the maximum extent for 5 seconds, hold for 5 seconds, revert head to neutral position.
• A 2 : stand for 5 seconds, bend the head backwards to the maximum extent for 5 seconds, hold for 5 seconds, revert head to neutral position.
• A 3 : stand for 5 seconds, flexion to the left to the maximum extent for 5 seconds, hold for 5 seconds, revert head to neutral position.
• A 4 : stand for 5 seconds, flexion to the right to the maximum extent for 5 seconds, hold for 5 seconds, revert head to neutral position.
• A 5 : stand for 5 seconds, rotate to the left to the maximum extent for 5 seconds, hold for 5 seconds, revert head to neutral position.
• A 6 : stand for 5 seconds, rotate to the right to the maximum extent for 5 seconds, hold for 5 seconds, revert head to neutral position.
• A 7 : stand for 5 seconds, raise hands to the maximum extent for 5 seconds, hold for 5 seconds, put hands on both sides of the body. Each movement is held for 20 seconds and repeated 3 times with 5 seconds rest between each repetition. In the data acquisition process, subject keeps still below shoulder. Finally, the sEMG activity from the upper trapezius (UT), cervical erector spinae (CE) and the sternocleidomastoid (SCM) are recorded bilaterally with Ag/AgCl surface electrodes. And the sEMG signal is converted to digital signal and sent to the laptop by the sEMG device.
Feature extraction The 5 kinds of common feature extraction methods, which include timedomain, frequency-domain, time-frequency, parametric model and nonlinear feature analysis, are used to extract features from the sEMG signal. The methods of using the disease-related knowledge to build features are also considered. We respectively extract 63 common features from the The FRR, DU, ACI, UN, SYM are extracted. The flexion relaxation ratio (FRR) is a useful, reliable marker to show altered neuromuscular function in both chronic neck pain patients and controls 14, 15 and can only be extracted from the S 1 (the data from A 1 ). We refer to the calculation method of the F RR in the paper 16 and extracted the FRR from S 1 . To our best knowledge, the DU, ACI, UN, SYM are firstly proposed to identify CS in our paper. According to formula 8, the DU is computed as the duration of muscle activation inspired by that the flexion relaxation phenomenon (FRP) which refers to a reduced or sudden onset of myoelectric silence in erector spinae muscles during full trunk flexion 17 . As shown in formula 9-12, we build new feature ACI based on that population with neck pain have an altered pattern of muscle activation in the deep and superficial cervical flexor muscles 18,19 . Besides, inspired by the imbalanced index proposed by Oddsson 20 and cross syndrome, the UN that include RMS ratio between different muscles was put forward in our work as shown in formula 13-22 20 . In order to explore the characteristics of bilateral muscle movement patterns, The SYM are extracted by calculating the similarity of the raw sEMG signal between symmetrical muscles using European distance, shape-based distance(SBD) 21   (2) (4)      X X X X X X X X X X X X X    X X X X X X X X X X X X X