A Novel Posture Based Approach To Fall Detection

Broadley R, Thies S, Kenney L, Klenk J, Granat M

5th ICAMPAM, 21 – 23 June 2017, Bethesda; Abstract Book Poster 1-51 p 61

BACKGROUND With a growing number of older adults, falls and their associated consequences are an increasing global health problem. It is widely accepted that there is an urgent need for improved automatic detection of falls, as current technology performs poorly and uptake is low. This study introduces a novel posture-based approach to fall detection. METHODS An algorithm was developed to detect falls through monitoring of changes in posture. The design was based on the premise that older adults would most commonly lie when in bed, and would pause in a seated posture as they got into bed. Therefore, any transition from upright to lying without a ‘reasonable’ period of being in a seated posture would signify a fall. The algorithm had one adjustable parameter, S, the length of time spent sitting before changing to a lying posture. This value was used to discriminate falls from controlled transitions. The algorithm’s performance was evaluated using two datasets; 144 simulated falls carried out by 8 healthy adults and 171 days of free-living data from 99 patients in a geriatric rehabilitation ward. The latter dataset contained no known falls. Data were collected using an activPAL3 attached to the thigh. 8- fold cross validation was used to optimise S and test sensitivity and false positive rate. S was optimised using Youden’s index for values of S from 0s to 5s in 0.1s increments. RESULTS Through 8-fold cross validation the optimised S was 1.44s 0.17 (mean SD). The mean sensitivity to detect the simulated falls was 96.5% 2.7. In the free-living data set 0.72 0.16 false positives were detected per day. CONCLUSIONS The use of posture change as a method to detect falls shows great potential and the low false positive rate is particularly encouraging. This approach needs to be tested using real-world falls.

10.1007/978-981-15-9897-5_8

Broadley R, Thies S, Kenney L, Klenk J, Granat M A Novel Posture Based Approach To Fall Detection 5th ICAMPAM, 21 – 23 June 2017, Bethesda; Abstract Book Poster 1-51 p 61 2017. https://ismpb.org/2017-bethesda.