Acceleration Magnitude at Impact Following Loss of Balance Can Be Estimated Using Deep Learning Model.

Pre-impact fall detection can detect a fall before a body segment hits the ground. When it is integrated with a protective system, it can directly prevent an injury due to hitting the ground. An impact acceleration peak magnitude is one of key measurement factors that can affect the severity of an injury. It can be used as a design parameter for wearable protective devices to prevent injuries. In our study, a novel method is proposed to predict an impact acceleration magnitude after loss of balance using a single inertial measurement unit (IMU) sensor and a sequential-based deep learning model. Twenty-four healthy participants participated in this study for fall experiments. Each participant worn a single IMU sensor on the waist to collect tri-axial accelerometer and angular velocity data. A deep learning method, bi-directional long short-term memory (LSTM) regression, is applied to predict a fall's impact acceleration magnitude prior to fall impact (a fall in five directions). To improve prediction performance, a data augmentation technique with increment of dataset is applied. Our proposed model showed a mean absolute percentage error (MAPE) of 6.69 ± 0.33% with r value of 0.93 when all three different types of data augmentation techniques are applied. Additionally, there was a significant reduction of MAPE by 45.2% when the number of training datasets was increased by 4-fold. These results show that impact acceleration magnitude can be used as an activation parameter for fall prevention such as in a wearable airbag system by optimizing deployment process to minimize fall injury in real time.

Machine Learning-Based Pre-Impact Fall Detection Model to Discriminate Various Types of Fall.

Pre-impact fall detection can send alarm service faster to reduce long-lie conditions and decrease the risk of hospitalization. Detecting various types of fall to determine the impact site or direction prior to impact is important because it increases the chance of decreasing the incidence or severity of fall-related injuries. In this study, a robust pre-impact fall detection model was developed to classify various activities and falls as multiclass and its performance was compared with the performance of previous developed models. Twelve healthy subjects participated in this study. All subjects were asked to place an inertial measuring unit module by fixing on a belt near the left iliac crest to collect accelerometer data for each activity. Our novel proposed model consists of feature calculation and infinite latent feature selection (ILFS) algorithm, auto labeling of activities, and application of machine learning classifiers for discrete and continuous time series data. Nine machine-learning classifiers were applied to detect falls prior to impact and derive final detection results by sorting the classifier. Our model showed the highest classification accuracy. Results for the proposed model that could classify as multiclass showed significantly higher average classification accuracy of 99.57 ± 0.01% for discrete data-based classifiers and 99.84 ± 0.02% for continuous time series-based classifiers than previous models (p < 0.01). In the future, multiclass pre-impact fall detection models can be applied to fall protector devices by detecting various activities for sending alerts or immediate feedback reactions to prevent falls.

A Novel Detection Model and Its Optimal Features to Classify Falls from Low- and High-Acceleration Activities of Daily Life Using an Insole Sensor System.

In order to overcome the current limitations in current threshold-based and machine learning-based fall detectors, an insole system and novel fall classification model were created. Because high-acceleration activities have a high risk for falls, and because of the potential damage that is associated with falls during high-acceleration activities, four low-acceleration activities, four high-acceleration activities, and eight types of high-acceleration falls were performed by twenty young male subjects. Encompassing a total of 800 falls and 320 min of activities of daily life (ADLs), the created Support Vector Machine model’s Leave-One-Out cross-validation provides a fall detection sensitivity (0.996), specificity (1.000), and accuracy (0.999). These classification results are similar or superior to other fall detection models in the literature, while also including high-acceleration ADLs to challenge the classification model, and simultaneously reducing the burden that is associated with wearable sensors and increasing user comfort by inserting the insole system into the shoe.

Predicting Complete Ground Reaction Forces and Moments During Gait With Insole Plantar Pressure Information Using a Wavelet Neural Network.

In general, three-dimensional ground reaction forces (GRFs) and ground reaction moments (GRMs) that occur during human gait are measured using a force plate, which are expensive and have spatial limitations. Therefore, we proposed a prediction model for GRFs and GRMs, which only uses plantar pressure information measured from insole pressure sensors with a wavelet neural network (WNN) and principal component analysis-mutual information (PCA-MI). For this, the prediction model estimated GRFs and GRMs with three different gait speeds (slow, normal, and fast groups) and healthy/pathological gait patterns (healthy and adolescent idiopathic scoliosis (AIS) groups). Model performance was validated using correlation coefficients (r) and the normalized root mean square error (NRMSE%) and was compared to the prediction accuracy of the previous methods using the same dataset. As a result, the performance of the GRF and GRM prediction model proposed in this study (slow group: r = 0.840-0.989 and NRMSE% = 10.693-15.894%; normal group: r = 0.847-0.988 and NRMSE% = 10.920-19.216%; fast group: r = 0.823-0.953 and NRMSE% = 12.009-20.182%; healthy group: r = 0.836-0.976 and NRMSE% = 12.920-18.088%; and AIS group: r = 0.917-0.993 and NRMSE% = 7.914-15.671%) was better than that of the prediction models suggested in previous studies for every group and component (p < 0.05 or 0.01). The results indicated that the proposed model has improved performance compared to previous prediction models.

Multi-class biological tissue classification based on a multi-classifier: Preliminary study of an automatic output power control for ultrasonic surgical units.

Ultrasonic surgical units (USUs) have the advantage of minimizing tissue damage during surgeries that require tissue dissection by reducing problems such as coagulation and unwanted carbonization, but the disadvantage of requiring manual adjustment of power output according to the target tissue. In order to overcome this limitation, it is necessary to determine the properties of in vivo tissues automatically. We propose a multi-classifier that can accurately classify tissues based on the unique impedance of each tissue. For this purpose, a multi-classifier was built based on single classifiers with high classification rates, and the classification accuracy of the proposed model was compared with that of single classifiers for various electrode types (Type-I: 6 mm invasive; Type-II: 3 mm invasive; Type-III: surface). The sensitivity and positive predictive value (PPV) of the multi-classifier by cross checks were determined. According to the 10-fold cross validation results, the classification accuracy of the proposed model was significantly higher (p<0.05 or <0.01) than that of existing single classifiers for all electrode types. In particular, the classification accuracy of the proposed model was highest when the 3mm invasive electrode (Type-II) was used (sensitivity=97.33-100.00%; PPV=96.71-100.00%). The results of this study are an important contribution to achieving automatic optimal output power adjustment of USUs according to the properties of individual tissues.