CC BY
0
SCIENCE TIME
COMPUTER VISION APPROACH IN ASSESSING HUMAN HEAD TREMOR
Meleshchenko Nikita Sergeevich, Zlobin Nikita Sergeevich, Savkov Vladimir Andreevich, Kataev Mikhail Yurievich, Tomsk State University of Control Systems and Radioelectronics, Tomsk,
E-mail: batnikov@internet.ru
Abstract. This paper explores the application of computer vision and machine learning technologies in medical research and diagnostics, particularly in the study of human head tremor as a motor disorder. It presents video analysis methods for detecting and quantitatively assessing head movements, which enhance diagnostics and monitoring of an individual's condition as the disease progresses. The developed software-hardware system allows individuals to perform monitoring at home, broadening the opportunity for timely response to changes. The results of the studies are presented, demonstrating the application of the developed methods in real medical conditions. This work underscores the importance of computer vision technologies in the advancement of modern medical diagnostics and provides prospects for further research in this field.
Key words: computer vision, imaging, machine learning, diagnostics, Parkinson's disease.
Motor disorders prevent many individuals from enjoying everyday life. As with other diseases, diagnostics and analysis are crucial in the treatment of such disorders. One progressive disorder that is important to assess and monitor is Parkinson's disease. The analysis of human movements in clinical medicine and therapy has been an evolving area of research since the early 2000s. Most existing human movement monitoring systems are linked to position determination systems using computer vision. These systems are based on the use of digital cameras, positioned in environments where uniform lighting can be ensured.
An important area of research related to movements involves the human head, whose motion characteristics serve as indicators of an individual's health status. Monitoring changes in head position during natural interactions with a digital camera allows for tracking head positions and detecting any changes. Furthermore, this approach enables observation of human movements in everyday home environments. Considering that classic tremor in Parkinson's disease, occurring at rest, is measured at frequencies from 4 to 7 Hz, with video at 30 frames per second, one 7 Hz oscillation
I
»
10
1 SCIENCE TIME 1
will have at least four frames, which is sufficient for confident detection of head position changes.
In this report, we propose an innovative solution that directly addresses the task of detecting head tremor and assessing its temporal and spatial characteristics. We suggest the development of an affordable and user-friendly motion capture system based on a digital camera, which can be used by a physician during examination or by an individual experiencing tremor in daily life. Regular assessments of a person's head tremor condition allow the characteristics of head movement to be made available to the physician. This quantitative approach, unlike visual observation, is objective, whereas visual is mostly subjective.
In this report, we propose an innovative solution that directly addresses the task of detecting head tremor and assessing its temporal and spatial characteristics. We propose building a convenient and affordable motion capture system based on a digital camera, which can be utilized by a physician during an examination or by an individual suffering from tremor in their daily life. If regular assessments of a person's head tremor condition are performed, information about the movement characteristics of the head will be available to the physician. This quantitative approach, unlike visual observation, is objective, whereas visual observation is largely subjective.
Typically, the assessment of a person's condition with Parkinson's disease is based on visual inspection and the physician's diagnosis. One of the characteristics assessed for the disease is the visible movement disorder of the head-tremor. We anticipate that computer vision and the mathematical interpretation of video movements of the head can provide a positive clinical outcome in assessing Parkinson's disease. Physicians typically assess the degree of motor disorders using the Parkinson's Disease Rating Scale (MDS-UPDRS [Goetz C.G. The Unified Parkinson's Disease Rating Scale (UPDRS): Status and Recommendations // Mov. Disord. - 2003. - No.7. - P.738-750.]). An experienced physician can recognize visual patterns that indicate the presence or severity of a specific disease. This is particularly true in neurology, as many neurological diseases affect human movement in a characteristic way, either reducing movement, altering it, or adding new movements.
In our view, computer interpretation of video will fundamentally enable an automated and objective assessment of visual features of neurological diseases, including the degree of head tremor. This does not require physicians to abandon the traditional approach to examination; rather, it provides them with a reliable, accurate quantitative assessment of the disease severity, as well as the ability to conduct comparisons of examinations over different periods of time and evaluate the trend of changes.
In the context of computer vision, the assessment of head position (pose) is the process of determining the orientation of the human head from digital RGB images. To achieve this, a specific sequence of processing steps is required to transform the pixel position of the head in the image. These steps include locating the human head in the image based on skin color and head shape, then segmenting the head in the image, and fixing the contour and center of gravity of the head. Assessing the head pose allows
1 SCIENCE TIME 1
for determining the orientation of the head relative to the global coordinate system. The head pose itself provides a rough indication of the gaze direction. When a person's eyes are clearly visible, the position of the head becomes essential for accurately predicting the head's position in space, and thus assessing changes in its position.
In clinical practice, there is often insufficient time to apply the full MDS-UPDRS scale, as it involves evaluation across more than 40 items, affecting various aspects such as head movement, gait, arm and leg movements, and voice. Typically, the time allocated for a patient visit allows for only the key aspects of the scale to be covered during an examination. In the clinical context of Parkinson's disease, an assessment of the severity of symptoms is required. Among all the elements of the scale, head tremor is one of the most common indicators used to determine the severity of the disease. The criteria for assessing the disease severity, according to the MDS-UPDRS scale, are as follows: 0. Normal: No tremor; 1. Very slight: Tremor occurs but does not interfere with any actions; 2. Slight: Tremor slightly interferes with certain actions; 3. Moderate: Tremor significantly interferes with many daily activities; 4. Severe: Tremor interferes with almost all activities. As evident from the assessment of disease severity, the complexity of the evaluation is related to the qualitative features of how the head tremor impacts daily activities. This highlights the need for precise and objective measurement tools that can quantify the tremor severity and its functional impact, providing a more systematic approach to symptom assessment in Parkinson's disease.
We propose a methodology based on capturing and processing video of a human head. The methodology relies on the following requirements: 1. Camera Placement: The camera should be positioned at eye level and directed straight at the subject's face to ensure a direct and unobstructed view. 2. Lighting: Avoid overexposure and shadows on the subject's face. Use diffused lighting to ensure uniform illumination without bright glares or deep shadows, which is crucial for accurate video analysis. 3. Background and Surroundings: The background behind the subject should be uniform and of a neutral color, free from any distracting objects such as flowers, paintings, wall clocks, or other people. This minimizes distractions and helps the algorithm focus on analyzing the tremor. 4. Clothing: The person should wear simple-cut clothing without complex patterns or bright colors. Dark-toned clothing is preferable to enhance the visibility of body contours. 5. Distance from the Camera: The subject should be at an arm's length from the camera. This ensures optimal frame filling and sufficient resolution for detailing the necessary movements. 6. Frame Composition: Only the head and upper part of the shoulders should be visible in the frame to focus attention maximally on head movements. This setup is designed to optimize the conditions for digital video analysis, enabling precise detection and analysis of head tremor, thus providing a robust tool for clinical assessment and research in neurological disorders.
In this report, we consider several methods for assessing the movement of the human head. The first method is based on creating an "energy portrait" of head movements, which involves calculating the probability of the head's position in a given pixel of the image. This approach has been employed in the research at the
I
SCIENCE TIME
I
Institute of Automation of the Chinese Academy of Sciences, where video analysis methods are used to study the CASIA-B dataset, specifically human gait. We have adapted this idea to evaluate human head movement (Figure 1).
А
Fig. 1 illustrates the energy portrait of head movement, comparing a person with tremor (A) and a healthy individual (B)
From Figure 1, we can clearly observe an "aura" associated with the head position during a tremor. By identifying the area with the highest probability of the head's position in a pixel (black color), we can calculate the distance covered by the "aura" (gray color). Considering that the distance between the person's head and the camera is, for example, 1 meter, and the camera has a resolution of 3000x2000 pixels, with a field of view of 90 degrees, the spatial resolution per pixel is 0.7 mm/pixel. If the "aura" spans 10 pixels, then the head tremor measures 7 mm. Now, with a physician's assessment of the tremor severity using the MDS-UPDRS scale, it is possible to correlate the measurements obtained from the video with the physician's evaluations to establish a clear quantitative scale that can be used by all physicians, as well as by individuals at home.
Another method for assessing head tremor involves measuring the rate of position change of a point attached to the person's nose. Figure 2 shows the velocities of head movements. The fluctuations of the point are typically characterized by changes in the horizontal (X) and vertical (Y) components. Visualizing the data of these point movements on the nose can reveal deviations in motor function, which is crucial in diagnosing and monitoring the condition of patients with Parkinson's disease and other disorders.
E
0 2S SO 7S 100 125 ISO 175 200
Frame
Fig. 2 Study of head tremor based on the movement of a point on the nose of a person with head tremor
1 SCIENCE TIME 1
Head Vetocity Over Time
- * Velocity
А аА 1 if м ч. i 1 i\n V \Л - У Velocity
•—лAr VW V U
1
1
0 75 50 75 100 125 150 175 300
Frame
Fig. 3 Study of head tremor using anthropometric points on the nose of a patient without head tremor
The comparison of Figures 2 and 3 demonstrates that the velocities of head movement in a person with Parkinson's disease differ from those of a healthy individual, which can be utilized for diagnostic purposes.
In conclusion, the two proposed methods allow for the acquisition of quantitative information about the state of a person with Parkinson's disease.
References:
1. Лихачев С.А., Ващилин В.В., Дик С.К. Тремор: феноменология и способы регистрации // Медицинский журнал: научно-практический рецензируемый журнал, 2010. - № 2. - С. 133-137.
2. Фролов С.В., Горбунов А.В., Потлов А.Ю. Регистрация и анализ тремора с помощью детектора движения на основе веб-камеры // Биомедицина. - 2012. -№ 2. - C. 80-83.
