KR-20260066602-A - METHOD FOR CLASSIFYING ABNORMAL GAIT IN PARKINSON'S PATIENTS BASED ON DEEP LEARNING MODEL MINIMIZING THE NUMBER OF SENSORS IN SMART INSOLE

Abstract

A method for detecting abnormal gait in Parkinson's patients based on a deep learning model capable of minimizing sensors in a smart insole is disclosed. The method includes the steps of: acquiring sensor data of a smart insole; and classifying abnormal gait patterns by analyzing the sensor data through a deep learning model based on a Convolutional Neural Network (CNN). The deep learning model may include a convolutional layer for distinguishing between the left foot and the right foot and extracting features from the sensor data.

Inventors

• 한창희
• 박은서

Assignees

• 고려대학교 세종산학협력단
• 주식회사 일리아스에이아이

Dates

Publication Date

20260512

Application Date

20250320

Priority Date

20241104

Claims (15)

1. A method for detecting abnormal gait in a computer device comprising at least one processor, A step of acquiring sensor data of a smart insole by the above-mentioned at least one processor; and A step of classifying abnormal walking patterns by analyzing sensor data through a CNN (Convolutional Neural Network)-based deep learning model by the above-mentioned at least one processor. Includes, The above deep learning model includes a convolutional layer for distinguishing between the left foot and the right foot and extracting features from the sensor data. An abnormal gait detection method characterized by
2. In paragraph 1, The above deep learning model includes a layer type that reflects temporal and spatial information as a convolutional layer for extracting features from the above sensor data. An abnormal gait detection method characterized by
3. In paragraph 1, The deep learning model above is composed of a combination of a first layer type that performs convolution operations in the temporal axis direction and a second layer type that performs convolution operations in the spatial axis direction, for a convolution layer to extract features from the sensor data. An abnormal gait detection method characterized by
4. In paragraph 3, The deep learning model above includes six convolutional layers as a combination of the first layer type and the second layer type. An abnormal gait detection method characterized by
5. In paragraph 4, The above six convolutional layers are arranged in the order of the above second layer type, the above first layer type, the above second layer type, the above first layer type, the above first layer type, and the above first layer type. An abnormal gait detection method characterized by
6. In paragraph 5, The second layer type used first among the above six convolutional layers analyzes each foot individually, and the second layer type used second analyzes both feet together. An abnormal gait detection method characterized by
7. In paragraph 3, The above deep learning model applies batch normalization to each convolutional layer to normalize the feature map. An abnormal gait detection method characterized by
8. In paragraph 3, The above deep learning model further includes a flatten layer for converting the feature map that has passed through the last convolutional layer into a one-dimensional vector, and a fully-connected layer for producing the final classification result for abnormal gait patterns. An abnormal gait detection method characterized by
9. In paragraph 1, The above classification step is, A step of extracting features by applying filters separately to the left foot and the right foot respectively through the above deep learning model. An abnormal gait detection method including
10. In paragraph 1, The above classification step is, Step of continuously detecting abnormal walking patterns using a moving window of a fixed time size An abnormal gait detection method including
11. At least one processor implemented to execute readable instructions on a computer device Includes, The above-mentioned at least one processor is, The process of classifying abnormal gait patterns by analyzing sensor data from smart insoles using a CNN (Convolutional Neural Network)-based deep learning model. Process, The above deep learning model includes a convolutional layer for distinguishing between the left foot and the right foot and extracting features from the sensor data. A computer device characterized by
12. In Paragraph 11, The deep learning model above is composed of a combination of a first layer type that performs convolution operations in the temporal axis direction and a second layer type that performs convolution operations in the spatial axis direction, for a convolution layer to extract features from the sensor data. A computer device characterized by
13. In Paragraph 12, The deep learning model above includes six convolutional layers as a combination of the first layer type and the second layer type. A computer device characterized by
14. In Paragraph 13, The above six convolutional layers are arranged in the order of the above second layer type, the above first layer type, the above second layer type, the above first layer type, the above first layer type, and the above first layer type. A computer device characterized by
15. In Paragraph 14, The second layer type used first among the above six convolutional layers analyzes each foot individually, and the second layer type used second analyzes both feet together. A computer device characterized by

Description

Method for Classifying Abnormal Gait in Parkinson's Patients Based on a Deep Learning Model Minimizing Sensors in Smart Insoles The following description concerns a technology for classifying abnormal gait patterns, such as those of Parkinson's disease patients. Walking is one of the representative human movements, and the analysis of walking patterns can be utilized in many application fields such as bioengineering, rehabilitation medicine, and healthcare. As a subfield of gait analysis, gait pattern classification is being researched for the diagnosis of Parkinson's disease, sports analysis, and the development of walking aids for the elderly, based on gait data acquired using sensors. Walking patterns are easily influenced by factors such as physical differences due to physical characteristics, speed, and terrain variations; these characteristics amplify variation within the same walking pattern, negatively impacting the performance of walking pattern classification. For example, even for the same walking motion, there are differences in patterns between walking on flat ground and walking on a hill, and these variations make it difficult to extract features to classify the walking form known as 'walking'. The gait classification system consists of a sensor module that acquires sensor data and an application module that calculates classification results based on the acquired data. Sensors used for gait classification mainly include video sensors, EMG (electromyographic) sensors, plantar pressure sensors, accelerometers, and gyroscopes. However, most sensors have limitations in that they can only measure walking data in restricted environments due to constraints such as sensor size and installation difficulties. Recent advancements in wearable sensor technology are leading to the lightweighting and simplification of equipment used for measuring walking data. Due to these factors, research on gait classification with relaxed location and behavioral constraints is actively underway, and many studies on gait classification using accelerometers and gyroscopes of smartphones and smartwatches are being conducted. Existing methods for analyzing gait patterns include camera video analysis systems, motion capture systems, and pressure mat systems. These conventional methods are characterized by the fact that they acquire and analyze gait signals in limited locations, such as hospitals or laboratories. Furthermore, they are characterized by high costs resulting from the use of high-performance cameras and motion analysis devices. To overcome the limitations of these existing gait analysis devices, research is being conducted to classify gait patterns using insoles attached inside shoes. Smart insole devices are being developed that enable the measurement of walking signals at a low cost without restrictions on location or time by incorporating pressure sensors, accelerometers, gyroscopes, and the like inside the insole. FIG. 1 is a block diagram illustrating an example of the internal configuration of a computer device in an embodiment of the present invention. FIG. 2 illustrates a deep learning model structure for classifying abnormal walking patterns in an embodiment of the present invention. Figure 3 illustrates an example of an experimental paradigm for constructing a smart insole dataset. FIG. 4 illustrates the process of selecting an ROI using a small number of sensors in a smart insole in one embodiment of the present invention. FIG. 5 illustrates an example of a preprocessing process used to classify abnormal gait by utilizing pressure sensor data of a smart insole in an embodiment of the present invention. FIG. 6 illustrates a model design that considers information regarding the left foot and the right foot separately in an embodiment of the present invention. Figure 7 illustrates the types of feature extraction layers of a convolutional neural network in one embodiment of the present invention. FIG. 8 illustrates a model architecture for a combination of different types of convolutional layers in an embodiment of the present invention. Figure 9 shows a unit step graph (S=normal person, EL=elderly, PD=Parkinson's disease patient) for comparing walking patterns between groups. FIGS. 10 to 12 illustrate the results of performance verification of a deep learning model for classifying abnormal walking patterns according to the present invention. Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings. Embodiments of the present invention relate to a technology for analyzing walking signals and classifying abnormal walking patterns, such as those of Parkinson's disease patients, using a smart insole device capable of acquiring human walking signals without restrictions on location and time. Embodiments including those specifically disclosed in this specification can implement an artificial intelligence model capable of diagnosing abnormal gait more a