Age-Related Reliability of B-Mode Analysis for Tailored Exosuit Assistance

Abstract

In the field of wearable robotics, assistance needs to be individualized for the user to maximize benefit. Information from muscle fascicles automatically recorded from brightness mode (B-mode) ultrasound has been used to design assistance profiles that are proportional to the estimated muscle force of young individuals. There is also a desire to develop similar strategies for older adults who may have age-altered physiology. This study introduces and validates a ResNet + 2x-LSTM model for extracting fascicle lengths in young and older adults. The labeling was generated in a semimanual manner for young (40,696 frames) and older adults (34,262 frames) depicting B-mode imaging of the medial gastrocnemius. First, the model was trained on young and tested on both young (R² = 0.85, RMSE = 2.36 ± 1.51 mm, MAPE = 3.6%, aaDF = 0.48 ± 1.1 mm) and older adults (R² = 0.53, RMSE = 4.7 ± 2.51 mm, MAPE = 5.19%, aaDF = 1.9 ± 1.39 mm). Then, the performances were trained across all ages (R² = 0.79, RMSE = 3.95 ± 2.51 mm, MAPE = 4.5%, aaDF = 0.67 ± 1.8 mm). Although age-related muscle loss affects the error of the tracking methodology compared to the young population, the absolute percentage error for individual fascicles leads to a small variation of 3-5%, suggesting that the error may be acceptable in the generation of assistive force profiles.

Keywords: aging; b-mode ultrasound; exoskeleton; fascicle length; muscle architecture; muscle dynamics; neural networks; wearable device.

Figure 1

The proposed study where an ultrasound probe is placed over the medial gastrocnemius muscles, the proposed ResNet + 2x-LSTM architecture extracts the fascicles, and then the performances of the model are evaluated across different walking speeds of 0.75, 1.25, and 1.5 m/s and tasks (level ground and incline).

Figure 2

Randomly selected representative ultrasound images for (A) older adults and (B) young adults at 0.75 m/s 0.0° level ground.

Figure 3

Illustration of high-level pipeline showing labeling used in this study. The continuous B-mode ultrasound recordings are segmented based on prerecorded heel strikes. The individual chucks are then processed using the affine optical flow algorithm of UltraTrack [25] to extract the fascicle lengths in millimeters (mm) across time in seconds (s) (in the lower left box). Each segmented sequence has many fascicle lengths (on average 130 frames in each segmented sequence), which form the fascicle lengths over time waveforms (left blue box). Finally, to validate the accuracy of the labeling methodology, the extracted individual segments of fascicles are plotted on top of each other over the stance phase (in the right box) following the approach proposed by Lai et al. [37].

Figure 4

Schematic of the proposed convolutional neural network with two long short-term memory (LSTM) units. The individual segmented sequences that are input to the ResNet model [35] are then processed by two layers of LSTM units. Finally, the sliding window method processed fixed, overlapped, chunked sequences, generating multiple predictions for each frame. Once the local minima and the local maxima are identified for overlapping sequences, the average prediction fascicle length is then reconstructed from each subsegment.

Figure 5

Fascicle lengths (in millimeters) versus time (in seconds) in output from the ResNet + 2x-LSTM algorithm for the all-age-trained model for both young and older adults compared against the (semiautomated) manual labeling. The time series data show the trends across three tasks, including level ground at walking speeds of 0.75 m/s (slower), 1.25 m/s at 10% incline, and 1.5 m/s (faster) of the muscle fascicles regressed from the B-mode ultrasound images.

Figure 6

Illustration of the mean absolute percentage error (MAPE) in percentage (%) for the fascicle lengths across participants (young in light green and older in light purple) across the four task velocities (level ground at walking speeds of 0.75, 1.25, and 1.5 m/s and 10% incline at 1.25 m/s). The results represent the training performed across participant ages using the ResNet + 2x-LSTM introduced in the study. The box plots extend from the lower (25th) to upper (75th) quartile values of the data, with a line at the median value. The whiskers (vertical lines) extend from the box to show the range of the data, from the minimum to the maximum value. Outliers (data points located outside the whiskers of the box plot and numerically distant from the rest of the data) are represented as black dots.