Limits...
Auto detection and segmentation of daily living activities during a Timed Up and Go task in people with Parkinson ’ s disease using multiple inertial sensors

View Article: PubMed Central - PubMed

ABSTRACT

Background: Wearable sensors have the potential to provide clinicians with access to motor performance of people with movement disorder as they undergo intervention. However, sensor data often have to be manually classified and segmented before they can be processed into clinical metrics. This process can be time consuming. We recently proposed detection and segmentation algorithms based on peak detection using Inertial Measurement Units (IMUs) to automatically identify and isolate common activities during daily living such as standing up, walking, turning, and sitting down. These algorithms were developed using a homogenous population of healthy older adults. The aim of this study was to investigate the transferability of these algorithms in people with Parkinson’s disease (PD).

Methods: A modified Timed Up And Go task was used since it is comprised of these activities, all performed in a continuous fashion. Twelve older adults diagnosed with early PD (Hoehn & Yahr ≤ 2) were recruited for the study and performed three trials of a 10 and 5-m TUG during OFF state. They were outfitted with 17 IMUs covering each body segment. Raw data from IMUs were detrended, normalized and filtered to reveal kinematics peaks that corresponded to different activities. Segmentation was accomplished by identifying the first minimum or maximum to the right and the left of these peaks. Segmentation times were compared to results from two examiners who visually segmented the activities. Specificity and sensitivity were used to evaluate the accuracy of the detection algorithms.

Results: Using the same IMUs and algorithms developed in the previous study, we were able to detect these activities with 97.6% sensitivity and 92.7% specificity (n = 432) in PD population. However, with modifications to the IMUs selection, we were able to detect these activities with 100% accuracy. Similarly, applying the same segmentation to PD population, we were able to isolate these activities within ~500 ms of the visual segmentation. Re-optimizing the filtering frequencies, we were able to reduce this difference to ~400 ms.

Conclusions: This study demonstrates the agility and transferability of using a system of IMUs to accurately detect and segment activities in daily living in people with movement disorders.

No MeSH data available.


Comparison between the original and modified algorithms for the standing up and sitting down. a Using the original algorithms that were previously developed using healthy older adults, we were able to achieve high specificity and sensitivity during both the 5 and 10 m TUG. However, the trunk and thigh IMU were prone to false positive (FP) due to sway in the trunk and thigh while participants were sitting in the chair or during pre-posturing by PD participants to gain leverage before standing up. b The modified algorithms used the orientation data from the thigh and sacrum modules to calculate the angle of the hip(θhip) to eliminate the FPs detected using the original algorithms. The angle was superimposed on the previous algorithms to eliminate the FPs and increase its accuracy to detect standing up and sitting down
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License 1 - License 2
getmorefigures.php?uid=PMC5384139&req=5

Fig4: Comparison between the original and modified algorithms for the standing up and sitting down. a Using the original algorithms that were previously developed using healthy older adults, we were able to achieve high specificity and sensitivity during both the 5 and 10 m TUG. However, the trunk and thigh IMU were prone to false positive (FP) due to sway in the trunk and thigh while participants were sitting in the chair or during pre-posturing by PD participants to gain leverage before standing up. b The modified algorithms used the orientation data from the thigh and sacrum modules to calculate the angle of the hip(θhip) to eliminate the FPs detected using the original algorithms. The angle was superimposed on the previous algorithms to eliminate the FPs and increase its accuracy to detect standing up and sitting down

Mentions: Several modifications to the algorithms and IMU selection were made to improve the detection of activities in PD patients (Table 2). These changes were made to enhance the detection algorithms by taking into account the biomechanics and movement strategies adopted by PD patients that were absent in healthy older adults. For example, during standing up and sitting down (Fig. 4), the angle of the hip (θhip) was added to further distinguished these activities from extraneous trunk and hip movements. θhip was calculated using the fused quaternion data of the sacrum and thigh. Furthermore, the band pass filter frequency of the trunk, which was re-optimized using the 10 m TUG data, was reduced from 1.57 Hz to 0.9 Hz to compensate for noise that might have been amplified by the tremors and postural instability. For walking, the sacrum IMU was replaced by the shin IMU while a new adaptive thresholding based on the histogram (numbers of bin = 20) of the signal amplitude (ay, Fig. 5b) was used to set the limit for task detection. This process is similar to Otsu’s thresholding [37] The threshold is defined as:Table 2


Auto detection and segmentation of daily living activities during a Timed Up and Go task in people with Parkinson ’ s disease using multiple inertial sensors
Comparison between the original and modified algorithms for the standing up and sitting down. a Using the original algorithms that were previously developed using healthy older adults, we were able to achieve high specificity and sensitivity during both the 5 and 10 m TUG. However, the trunk and thigh IMU were prone to false positive (FP) due to sway in the trunk and thigh while participants were sitting in the chair or during pre-posturing by PD participants to gain leverage before standing up. b The modified algorithms used the orientation data from the thigh and sacrum modules to calculate the angle of the hip(θhip) to eliminate the FPs detected using the original algorithms. The angle was superimposed on the previous algorithms to eliminate the FPs and increase its accuracy to detect standing up and sitting down
© Copyright Policy - OpenAccess
Related In: Results  -  Collection

License 1 - License 2
Show All Figures
getmorefigures.php?uid=PMC5384139&req=5

Fig4: Comparison between the original and modified algorithms for the standing up and sitting down. a Using the original algorithms that were previously developed using healthy older adults, we were able to achieve high specificity and sensitivity during both the 5 and 10 m TUG. However, the trunk and thigh IMU were prone to false positive (FP) due to sway in the trunk and thigh while participants were sitting in the chair or during pre-posturing by PD participants to gain leverage before standing up. b The modified algorithms used the orientation data from the thigh and sacrum modules to calculate the angle of the hip(θhip) to eliminate the FPs detected using the original algorithms. The angle was superimposed on the previous algorithms to eliminate the FPs and increase its accuracy to detect standing up and sitting down
Mentions: Several modifications to the algorithms and IMU selection were made to improve the detection of activities in PD patients (Table 2). These changes were made to enhance the detection algorithms by taking into account the biomechanics and movement strategies adopted by PD patients that were absent in healthy older adults. For example, during standing up and sitting down (Fig. 4), the angle of the hip (θhip) was added to further distinguished these activities from extraneous trunk and hip movements. θhip was calculated using the fused quaternion data of the sacrum and thigh. Furthermore, the band pass filter frequency of the trunk, which was re-optimized using the 10 m TUG data, was reduced from 1.57 Hz to 0.9 Hz to compensate for noise that might have been amplified by the tremors and postural instability. For walking, the sacrum IMU was replaced by the shin IMU while a new adaptive thresholding based on the histogram (numbers of bin = 20) of the signal amplitude (ay, Fig. 5b) was used to set the limit for task detection. This process is similar to Otsu’s thresholding [37] The threshold is defined as:Table 2

View Article: PubMed Central - PubMed

ABSTRACT

Background: Wearable sensors have the potential to provide clinicians with access to motor performance of people with movement disorder as they undergo intervention. However, sensor data often have to be manually classified and segmented before they can be processed into clinical metrics. This process can be time consuming. We recently proposed detection and segmentation algorithms based on peak detection using Inertial Measurement Units (IMUs) to automatically identify and isolate common activities during daily living such as standing up, walking, turning, and sitting down. These algorithms were developed using a homogenous population of healthy older adults. The aim of this study was to investigate the transferability of these algorithms in people with Parkinson’s disease (PD).

Methods: A modified Timed Up And Go task was used since it is comprised of these activities, all performed in a continuous fashion. Twelve older adults diagnosed with early PD (Hoehn & Yahr ≤ 2) were recruited for the study and performed three trials of a 10 and 5-m TUG during OFF state. They were outfitted with 17 IMUs covering each body segment. Raw data from IMUs were detrended, normalized and filtered to reveal kinematics peaks that corresponded to different activities. Segmentation was accomplished by identifying the first minimum or maximum to the right and the left of these peaks. Segmentation times were compared to results from two examiners who visually segmented the activities. Specificity and sensitivity were used to evaluate the accuracy of the detection algorithms.

Results: Using the same IMUs and algorithms developed in the previous study, we were able to detect these activities with 97.6% sensitivity and 92.7% specificity (n = 432) in PD population. However, with modifications to the IMUs selection, we were able to detect these activities with 100% accuracy. Similarly, applying the same segmentation to PD population, we were able to isolate these activities within ~500 ms of the visual segmentation. Re-optimizing the filtering frequencies, we were able to reduce this difference to ~400 ms.

Conclusions: This study demonstrates the agility and transferability of using a system of IMUs to accurately detect and segment activities in daily living in people with movement disorders.

No MeSH data available.