Journal: Sensors (Basel, Switzerland)

Article Title: Augmenting a ResNet + BiLSTM Deep Learning Model with Clinical Mobility Data Helps Outperform a Heuristic Frequency-Based Model for Walking Bout Segmentation

doi: 10.3390/s25206318

Figure Lengend Snippet: Sensor placements for PAMAP2 ( a ), healthy dataset ( b ), and clinical dataset ( c ). The PAMAP2 sensor was attached to the dominant side lateral ankle with a strap. The healthy dataset was attached using a semi-elastic strap, with the sensor placed medially and inferior to the knee. The clinical dataset used the same attachment location as the healthy dataset, but the sensors were attached using medical-grade tape. Please note that the IMUs are not to scale, as increased size is used to clearly show placements.

Article Snippet: Finally, the clinical dataset included 32 older adults with OA awaiting knee (n = 25) or hip (n = 7) arthroplasty (12M, 20F; 65.7 ± 7.8 yr; 168.0 ± 9.9 cm; 92.3 ± 23.9 kg), who wore bilateral tibial-mounted IMUs (Axivity AX6, 100 Hz) during one week of free-living and in-clinic assessments; only quiet standing and 60 s self-selected walking tasks during the in-clinic assessments were labeled, with gait speeds stratified as slow (<0.8 m/s), average (0.8–1.2 m/s), or fast (>1.2 m/s).

Techniques: