In this research, we developed a new CM classification method focusing on drawing actions using a commercially available tablet terminal and a stylus pen. Drawing time and drawing pressure were recorded while participants traced spiral, square, and triangular waves on a tablet. Using these data and machine learning algorithms, we demonstrated high classification accuracy with 76% sensitivity, 76% specificity, and AUC 0.80. A previous report showed a sensitivity of 61% to 61.7% and an AUC of 0.74 for the 10-second grip and release test.12, 13, 14the finger escape sign showed a sensitivity of 48–55%.12, 13deep tendon reflex changes showed sensitivities of 15% to 56%.15,16. Our method is more sensitive than conventional physical testing, demonstrating its utility as a screening tool.
There are many reports of handwriting in neurological disorders17, 18, 19. Previous studies have examined pen pressure and kinematic features of spiral painting in patients with Parkinson’s disease using a diagnostic that assesses the severity of movement disorders in patients with other neurodegenerative diseases. is also used to20, 21, 22, 23. A machine learning-based composite diagnostic tool for Parkinson’s disease has also been developed and demonstrated high accuracy.24,25. Furthermore, a high-accuracy classifier for Alzheimer’s disease has been reported, focusing on features related to the speed of handwritten signatures.26. There are also some reports on how to detect dysgraphia in children.27,28. A tablet-based diagnostic tool has been developed to predict writing disorders by focusing on the static, kinematic, pressure, and tilt characteristics of writing motion.29. However, to our knowledge, no studies have been reported on how to diagnose writing disorders in CM using machine learning. This research is the first report of a simple CM review method based on drawing tasks using machine learning.
In CM, the intrinsic muscle function of the hand is reduced.30 and there is obvious spasticity of the hands31, which can cause clumsiness. Therefore, it is clinically plausible that the mean drawing pressure would be lower overall. Regarding the writing pressure smoothness indicated by the spectral arc length (SPARC), the CM group showed significantly lower spiral wave values and lower smoothness than the non-CM group, which is also a reasonable finding. is. On the other hand, the square wave and triangular wave have larger values in both groups than the spiral, and it was found that these shapes have a smoother pen pressure. However, his SPARC values for square and triangle waves did not differ significantly between groups. The reason is that these two figures consist of several short lines with short pauses in between, making it more difficult to distinguish smoothness compared to spirals written continuously over a long period of time. This is probably because On the other hand, there was no significant difference in drawing time for any shape. Participants were asked to write at their preferred speed, with no instructions as to how fast they should write. Therefore, in every group, some people write quickly, while others write slowly and carefully. Different results may be observed when instructed to write as quickly as possible, and further studies are needed to verify this. For the SVM classification model, higher accuracy was observed with the model using the triangular wave. This is likely because the triangular wave requires the tip of the stylus to return in the opposite direction, which is difficult for his CM patient who has poor hand control. In addition, previous reports investigating writing angles found that right-handers have biomechanical properties in their hand muscles that favor upward-sloping lines and prevent downward-sloping lines.32,33. Studies have also reported differences in speed and line length accuracy when older people draw upward and downward lines.34. In this study, all participants were right-handed, and a rightward triangular wave containing both rising and falling lines may have made the difference between CM patients more pronounced, distinguishing CM from non-CM. Differentiating accuracy may have improved. .
The method introduced in this report does not require any special equipment other than a commercially available tablet terminal and stylus pen. It can be introduced not only in hospitals but also outside hospitals such as at home. Although this is a cross-sectional study in patients with CM before diagnosis and further prospective studies are needed, this method may lead to techniques for early detection of CM, allowing patients to confirm the diagnosis and initiate early treatment. It may prompt you to visit a spine specialist to receive it. There are several approaches for early detection of CM outside the hospital. For example, web-based symptom checkers are widely used diagnostic tools for CM. However, these tools are limited in their ability to accurately diagnose mild symptoms and require further optimization.35. Previous studies have reported a CM diagnostic system that uses non-contact sensor devices and artificial intelligence to analyze handgrip and release, which may lead to early detection out of hospital.36. Our study is novel in that it objectively analyzed writing behavior, a common activity in daily life. At present, it is necessary to use a specific shape, but in the future this method can also be applied to unspecified shapes, such as when writing a name, and it may lead to the development of a method that allows unconscious screening in everyday life. There is
This study has some limitations. First, we did not analyze other conditions that affect writing, such as carpal tunnel syndrome, cubital tunnel syndrome, and Parkinson’s disease. We have already developed a taxonomy for carpal tunnel syndrome and will classify multiple disorders in future studies. Second, the CM group included only preoperative patients, raising concerns about a higher rate of severe cases. However, the fact that we included cases classified as mild and moderate using the Japanese Orthopedic Association (JOA) scoring system suggests that this method could be used as a screening tool. Third, we did not limit the level of spinal cord compression or conduct a comparison between levels. The main purpose of this method is screening, and the correlation between the degree of stenosis and the degree of impairment in writing needs to be investigated separately. Fourth, the method is currently not sensitive enough to be used as a screening tool. Although higher than traditional health checks, greater sensitivity and accuracy are required for out-of-hospital and everyday use. However, the strength of this model is that it can be updated sequentially by adding cases, and we aim to get closer to implementation by improving accuracy in the future. For example, adding features related to sensation and muscle strength can improve model accuracy. These will be added to the analysis as the number of samples increases in the future.
In conclusion, we have developed a novel classification method that provides a basis for CM screening systems using analysis of drawing behavior based on machine learning algorithms. By integrating features related to drawing behavior, we obtained a model with high classification accuracy. This method can be used to develop an in-hospital or out-of-hospital disease screening system to facilitate early detection and treatment of CM using only a commercially available tablet and stylus.
