Ifsttar PhD subject

French version
Detailed form :

Title : Estimation of walking characteristics using low-cost, non-dedicated sensors for the early detection of the risk of falling in the elderly.

Main host Laboratory - Referent Advisor   -

Director of the main host Laboratory   -

PhD Speciality biomécanique, traitement du signal

Axis of the performance contract 1 - COP2017 - Efficient transport and safe travel

Main location Bron

Doctoral affiliation UNIVERSITE CLAUDE-BERNARD-LYON 1

PhD school MEGA (MECANIQUE, ENERGETIQUE, GENIE CIVIL, ACOUSTIQUE)

Planned PhD supervisor RENAUDIN Valérie  -  Université Gustave Eiffel  -  AME - GEOLOC

Planned PhD co-supervisor ROBERT Thomas  -  Université Gustave Eiffel  -  TS2 - LBMC

Planned financing Contrat doctoral  - Ifsttar

Abstract

Falls in older adults have great medical, social and financial impact. The aim of this thesis is to propose a ubiquitous method to extract relevant fall risk parameters from reallife inertial data regardless of device type and placement.

The main difficulty is estimating fall risk parameters independent of sensor placement (wrist, pocket, etc.). For this, we propose to calculate fall risk parameters based on discrete step time series. Then, the problem is transferred to finding a robust step detection algorithm which we developed and validated against different populations, placements and activities. Next, we calculated fall risk parameters and tested their association with future falls on an ambulatory dataset (one week of recording) of 300 elderly people.

The step detection method had an average precision and recall of 99% and 95% respectively when evaluated on datasets with different populations (young, elderly, blind), walking conditions (outdoors, using walking aid), and sensor placements (jacket, pants pocket, handheld). The association between calculated parameters and prospective falls had an AUC of 0.7 if parameters are aggregated on walking bouts greater than 200 steps (2-minutes). This AUC is comparable to models made with a fixed sensor placement. However, selecting long walking bouts causes the exclusion of participants who do not walk long enough (6% of population considered).

The proposed novel solution is ubiquitous and can reach the broad public. It can open doors toward personal monitoring of fall risk status using consumer devices.

Keywords : Fall risk; Step detection; Consumer devices; Inertial Measurement Units; Fall prediction; Sensor placement; Older adults

Main host Laboratory - Referent Advisor	-
Director of the main host Laboratory	-
PhD Speciality	biomécanique, traitement du signal
Axis of the performance contract	1 - COP2017 - Efficient transport and safe travel
Main location	Bron
Doctoral affiliation	UNIVERSITE CLAUDE-BERNARD-LYON 1
PhD school	MEGA (MECANIQUE, ENERGETIQUE, GENIE CIVIL, ACOUSTIQUE)
Planned PhD supervisor	RENAUDIN Valérie - Université Gustave Eiffel - AME - GEOLOC
Planned PhD co-supervisor	ROBERT Thomas - Université Gustave Eiffel - TS2 - LBMC
Planned financing	Contrat doctoral - Ifsttar
Abstract Falls in older adults have great medical, social and financial impact. The aim of this thesis is to propose a ubiquitous method to extract relevant fall risk parameters from reallife inertial data regardless of device type and placement. The main difficulty is estimating fall risk parameters independent of sensor placement (wrist, pocket, etc.). For this, we propose to calculate fall risk parameters based on discrete step time series. Then, the problem is transferred to finding a robust step detection algorithm which we developed and validated against different populations, placements and activities. Next, we calculated fall risk parameters and tested their association with future falls on an ambulatory dataset (one week of recording) of 300 elderly people. The step detection method had an average precision and recall of 99% and 95% respectively when evaluated on datasets with different populations (young, elderly, blind), walking conditions (outdoors, using walking aid), and sensor placements (jacket, pants pocket, handheld). The association between calculated parameters and prospective falls had an AUC of 0.7 if parameters are aggregated on walking bouts greater than 200 steps (2-minutes). This AUC is comparable to models made with a fixed sensor placement. However, selecting long walking bouts causes the exclusion of participants who do not walk long enough (6% of population considered). The proposed novel solution is ubiquitous and can reach the broad public. It can open doors toward personal monitoring of fall risk status using consumer devices.
Keywords :	Fall risk; Step detection; Consumer devices; Inertial Measurement Units; Fall prediction; Sensor placement; Older adults

List of topics

Applications closed