=Paper=
{{Paper
|id=Vol-2820/paper5
|storemode=property
|title=Detecting Signatures of Early-stage Dementia with
Behavioural Models Derived from Sensor Data
|pdfUrl=https://ceur-ws.org/Vol-2820/AAI4H-5.pdf
|volume=Vol-2820
|authors=Rafael Poyiadzi,Weisong Yang,Yoav Ben-Shlomo,Ian Craddock,Liz Coulthard,Raul Santos-Rodriguez,James Selwood,Niall Twomey
|dblpUrl=https://dblp.org/rec/conf/ecai/PoyiadziYBCCSST20
}}
==Detecting Signatures of Early-stage Dementia with
Behavioural Models Derived from Sensor Data==
https://ceur-ws.org/Vol-2820/AAI4H-5.pdf
Detecting Signatures of Early-stage Dementia with
Behavioural Models Derived from Sensor Data
Rafael Poyiadzi*1 and Weisong Yang*2 and Yoav Ben-Shlomo 3 and Ian Craddock 4
Liz Coulthard 5 and Raul Santos-Rodriguez 6 and James Selwood7 and Niall Twomey 8
Abstract. There is a pressing need to automatically understand the adequate temporal resolution to understand disease progression and
state and progression of chronic neurological diseases such as de- deterioration since AD has high variance across patients.
mentia. The emergence of state-of-the-art sensing platforms offers Some symptoms are prevalent across dementias. Confusion can
unprecedented opportunities for indirect and automatic evaluation of lead to unpredictable behaviour and activity abandonment during
disease state through the lens of behavioural monitoring. This paper normal daily routines. Interestingly, behavioural interactions between
specifically seeks to characterise behavioural signatures of mild cog- persons with dementia (PwD) and their carers/family are also affected
nitive impairment (MCI) and Alzheimer’s disease (AD) in the early by disease progression and efforts by PwD to allay confusion and
stages of the disease. We introduce bespoke behavioural models and uncertainty often result in them persistently seeking the company of
analyses of key symptoms and deploy these on a novel dataset of persons whom they trust. In this work we intend to address shadowing
longitudinal sensor data from persons with MCI and AD. We present [23] (persistently staying in the company of a trusted carer/family),
preliminary findings that show the relationship between levels of sleep wandering [24] (going from room to room with nonspecific intent)
quality and wandering can be subtly different between patients in the and disturbed sleep [15] (a common symptom for PwD).
early stages of dementia and healthy cohabiting controls. Progression of dementia and MCI are characterised by an increase
in symptoms that will affect normal daily activity and behaviour. It is
important to remember that these symptoms affect normal daily activ-
1 Introduction ity and behaviour. In understanding the dependence between normal
behaviour and the expression of symptoms, an opportunity to under-
Dementia is a progressive neurological condition affecting cognition stand and quantify disease state indirectly via behavioural change is
and behaviour with a significant impact on activities of daily living. It exposed. Modern sensing technology offers promise in monitoring
is one of the major causes of disability and dependency among elderly daily behaviour, extracting symptom expression rates and measure-
population with approximately 50 million people with the disease ment of state and progression of the disease [7, 30]. In contrast to
world wide. Alzheimer’s disease (AD) is the most common cause of normal clinical evaluation that occur in foreign environments, ours is
dementia. Mild cognitive impairment (MCI) refers to a decline that achieved from the patients’ residences.
has minimal impact on activities of daily living [13]. Patients with This paper outlines novel computational behavioural analysis al-
MCI can continue to live independently [26]. Not all patients with gorithms for modelling disease state and progression. The potential
MCI will convert to dementia, but they are at increased risk. MCI has for detecting unseen behavioural bio-markers of the early stages of
an annual conversion rate to AD of between 5 and 10% [27]. dementia is exposed by analysis on the longitudinal, in-home sensor
Making an accurate, reliable diagnosis of dementia is a challenge. data. Section 2 reviews prior work on smart homes and dementia. Our
Patients may find cognitive tests stressful, which impacts their perfor- data collection and modelling procedures are outlined in Section 3.
mance. This may be exacerbated in a new, unknown, clinical environ- Results are presented in Section 4 and we conclude in Section 5.
ment with an unfamiliar clinician. Performance on cognitive tests also
does not give an indication of how someone is managing in the real
world. Although diagnosis may include bio-markers measurement, 2 Related Work
in addition to this being invasive and expensive, these do not have
Numerous smart systems have been designed and developed to moni-
1 University of Bristol, England, email: rp13102@bristol.ac.uk
tor the well-being and health status of elders, such as the GATOR Tech
2 University of Bristol, England, email: ws.yang@bristol.ac.uk
Smart House [14], the AWARE home [1], the Microsoft’s EasyLiving
3 University of Bristol, England, email: y.ben-shlomo@bristol.ac.uk
4 University of Bristol, England, email: ian.craddock@bristol.ac.uk
project [4] and the MavHome Project [6]. In the dementia domain,
5 University of Bristol, England, email: elizabeth.coulthard@bristol.ac.uk [22] introduced a support system to aid doctors diagnosing dementia.
6 University of Bristol, England, email: enrsr@bristol.ac.uk Participants only needed to perform a selection of Instrumental Activ-
7 University of Bristol, England, email: james.selwood@bristol.ac.uk ities of Daily Living (IADL) in a smart home environment. However,
8 Cookpad Ltd, and University of Bristol, England (honorary), email: niall- the main limitation is that this approach relies on data collected in a
twomey@cookpad.com laboratory environment and not in a real world setting.
* Authors contributed equally
Along these lines, [3] undertook a study on data acquired by The
Copyright © 2020 for this paper by its authors. Use permitted under Creative
Commons License Attribution 4.0 International (CC BY 4.0). This volume is
ORegon Centre for Aging and TECHnology (ORCATECH) at the
published and copyrighted by its editors. Advances in Artificial Intelligence Oregon Health and Science University, that investigated the walking
for Healthcare, September 4, 2020, Virtual Workshop. speed and general activity in the homes of participants in order to
detect dementia. The extension presented in [2] modelled the pres- mathematical terms we assume that training and testing distribu-
ence of the participants in different rooms using Poisson processes to tions differ, i.e. 𝑃𝑡𝑟 (𝑥) ≠ 𝑃𝑡𝑒 (𝑥), but that conditional distributions
demonstrate statistical differences between different states of cogni- remain the same 𝑃𝑡𝑟 (𝑦|𝑥) = 𝑃𝑡𝑒 (𝑦|𝑥). MMD offers an elegant ap-
tion. Localisation has been found to be a common ground for different proach to quantify and match distributional shift and we follow the
works. For instance [18] and [29] use an Ultra Wideband device to try methodology outlined by [17] in our work.
to measure wandering behaviours. However, Radio Frequency (RF)
identification may be limited by its detection range since unpowered
3.2.2 Semi-Supervised Learning
RF devices operate only over a meter’s distance. Also, [10] uses a
relatively complex sensor network consisted of RF tracking and mo- Since only a small portion of labelled data are available, we leverage
tion and heading sensors for monitoring and studying behavioural supervised (L 𝑆𝐿 ) semi-supervised (L 𝑆𝑆𝐿 ) objectives to make effi-
patterns of patients with dementia. cient use of data. The supervised objective (L 𝑆𝐿 ) is the traditional
We build upon the existing literature by focusing on data collected conditional log-likelihood that one optimises for in Conditional Ran-
in the wild that, while more challenging to analyse, also provides dom Fields (CRF) [31]. The data and labels for this are obtained from
a richer and pervasive perspective on the subjects’ behaviour. Addi- the walkthrough that was introduced in the previous section. CRF
tionally, our aim is to automatically detect behavioural symptoms that model sequence probabilities as:
may suggest early dementia. Finally, we want go beyond the analysis !
𝑇
of a single participant and incorporate the monitoring of interactions −1
Õ
𝑃 𝜃 0 ( 𝒚| 𝑿) = 𝑍 ( 𝑿) · exp 𝜙𝑡 (𝑥 𝑡 , 𝑦 𝑡 , 𝑦 𝑡−1 ) (1)
within the home environment as a key element to understand the
𝑡=1
progression of the disease.
where 𝜙𝑡 are the log-potentials and 𝑍 ( 𝑿) is the partition function. In
3 Methods our case the log-potentials decompose, to emission function and tran-
sition function, as follows: 𝜙𝑡 (𝑥 𝑡 , 𝑦 𝑡 , 𝑦 𝑡−1 ) = 𝑢 (𝑥 𝑡 , 𝑦 𝑡 ) · 𝜏(𝑦
˜ 𝑡 , 𝑦 𝑡−1 ),
This section introduces our data collection and modelling pipelines. where:
3.1 Data collection
𝜏(𝑦 𝑡 , 𝑦 𝑡−1 ), 𝛼(𝑡) >= 𝜖 0
˜ 𝑡 , 𝑦 𝑡−1 ) =
𝜏(𝑦 (2)
Ethical approval was secured from the Carmarthen Research Ethics I 𝑦𝑡 =𝑦𝑡−1 , otherwise
Committee and the Health Research Authority (HRA) to record sensor where 𝛼(𝑡) is the activity level of the participant at time 𝑡 (calculated
data from the homes of people who have received a diagnosis of as the average absolute jerk of the tri-axial signals) and 𝜖 0 denotes a
dementia or have MCI for up to 12 months using the SPHERE [36, threshold, below which we assume there is no movement. This forbids
35, 8] system. Informed consent was obtained to analyse this data to location transitions unless a certain activity potential is surpassed.
determine links between behavioural patterns, residents of the home The emission potentials are modelled with a 4-layer fully-
and disease state. Participants are recruited from North Bristol NHS connected neural network, 𝑢(𝑥, 𝑦) = ℎ(𝑥) 𝑦 , i.e. the 𝑦-th element
Trust, the Dementia Wellbeing Service in Bristol and the Research of ℎ(𝑥). The semi-supervised objective (L 𝑆𝑆𝐿 ) constructed by itera-
Institute for the Care of Older People (RICE) Centre in Bath. The tively using the most likely label sequence as temporary targets, i.e.
SPHERE sensor system is installed into their homes [36, 35]. 𝒚 ∗ = arg max𝒚 ∈Y 𝑃(𝒚| 𝑿) [28].
The key sensors that we use in the analysis of this work are ac-
celerometers [11, 12, 34]. These are set record at a rate of 20 Hz. This
data is broadcast to gateways by means of radio links. Six gateways are 4 Findings
installed in every house. These simultaneously record accelerometer In this section we present findings relevant to sleeping disturbance,
readings and the Received Signal Strength Intensity (RSSI). wandering and shadowing. We illustrate the analysis using 2 different
In-home localisation is a key task in this work that leverages the houses that are inhabited by either a PwD, or by someone who has
RSSI data [20, 5, 19]. Labelled data are acquired from a ‘walkthrough’ been founds to have MCI, and one partner. Data collection is ongoing,
script that is performed by the technician upon installation. Each room so these are preliminary findings and some houses have significantly
is visited while holding the wearables and RSSI and wearable data more data than others. The analyses below result from localisation
are simultaneously recorded. The real-time location of the technician and activity level predictions that were trained and validated on the
is recorded with an annotation app. walkthrough data.
To regress from sensor data to localisation and activity predictions,
we extract basic features from windowed data (5 seconds length with
overlap of 2.50 seconds following [33]). Features include: mean, stan- 4.1 Shadowing
dard deviation (std), max, min, diff (first-order difference operator) The mutual information (MI) between the location of the two resi-
and the count of missing values. dents in each house is measured to estimate shadowing, following a
We can complement labelled data by introducing domain knowl- similar approach to that described in [32]. Additionally, in our analy-
edge into the analysis by suggesting reasonable locations based on sis we stratify MI by time of the day in order to understand whether
calculable context. For example, if predicting ‘sleep’ at 3AM it is significant MI is confined to particular intervals, e.g. morning Fig.1.
likely that the resident is in the bedroom. Temporal lags were introduced in order to allow for short delays in
shadowing, with no significant change, and are therefore not included
3.2 Modelling considerations in this work. This is likely due to the fact that the lag interval is
significantly shorter than the analysis window. Looking at Fig.1, in the
3.2.1 Covariate Shift
case of house B there is no trend of increasing/decreasing correlation.
We account for our expectations of covariate shift (due to disease Even though in the case of house A, there is a slight increase over
progression) with Maximum Mean Discrepancy (MMD) [17]. In time, stronger conclusions will be drawn when more data arrives.
14
10 12
8 10
6 8
4 6
4
2 2
0 0
10 14
12
8 10
6 8
4 6
4
2 2
0 0
9 0 1 2 3 4 5 6 7 1 5 9 1 5 9 3 7 1
9-0
7-1
9-0
7-2
9-0
7-2
9-0
7-2
9-0
7-2
9-0
7-2
9-0
7-2
9-0
7-2
9-0
7-2 9-2 9-2 9-2 0-0 0-0 0-0 0-1 0-1 0-2
201 201 201 201 201 201 201 201 201 9-0 9-0 9-0 9-1 9-1 9-1 9-1 9-1 9-1
201 201 201 201 201 201 201 201 201
00:00:00- 08:00:01- 14:00:01- 20:00:01- 00:00:00- 08:00:01- 14:00:01- 20:00:01-
06:00:00 10:00:00 18:00:00 22:00:00 06:00:00 10:00:00 18:00:00 22:00:00
06:00:01- 10:00:01- 18:00:01- 06:00:01- 10:00:01- 18:00:01-
08:00:00 14:00:00 20:00:00 08:00:00 14:00:00 20:00:00
(a) House A (b) House B
Figure 1: MI derived for wandering, separated in different segments of the day.
120
100
80
00:00:00-
06:00:00
60
40
20
0
400
300
06:00:01-
10:00:00
200
100
0
8 9 0 1 2 3 4 5 6 7 8 8 9 0 1 2 3 4 5 6 7 8
7-1 7-1 7-2 7-2 7-2 7-2 7-2 7-2 7-2 7-2 7-27-1 7-1 7-2 7-2 7-2 7-2 7-2 7-2 7-2 7-2 7-2
9-0 9-0 9-0 9-0 9-0 9-0 9-0 9-0 9-0 9-0 9-09-0 9-0 9-0 9-0 9-0 9-0 9-0 9-0 9-0 9-0 9-0
201 201 201 201 201 201 201 201 201 201 201201 201 201 201 201 201 201 201 201 201 201
MCI Control
bedroom 1 kitchen 1 total
hall 1 living room 1
(a) House A
140
120
100
00:00:00-
06:00:00
80
60
40
20
0
500
400
06:00:01-
300
10:00:00
200
100
0
1 5 9 1 5 9 3 7 1 1 5 9 1 5 9 3 7 1
9-2 9-2 9-2 0-0 0-0 0-0 0-1 0-1 0-2 9-2 9-2 9-2 0-0 0-0 0-0 0-1 0-1 0-2
9-0 9-0 9-0 9-1 9-1 9-1 9-1 9-1 9-1 019-0 019-0 0190-019-1 019-1 019-1 019-1 019-1 019-1
201 201 201201 201 201 201 201 201 2 2 2 2 2 2 2 2 2
Dementia Control
bedroom 2 kitchen 1 total
dining room 1 living room 1
(b) House B
Figure 2: The total activity per room for House A (upper) and House B (lower).
0.030 0.05
0.025 0.04
00:00:00-
00:00:00-
06:00:00
06:00:00
0.020 0.03
0.015 0.02
0.010
0.01
0.005
0.00
0.030 0.05
0.025 0.04
06:00:01-
06:00:01-
10:00:00
10:00:00
0.020 0.03
0.015 0.02
0.010
0.01
0.005
0.00
7- 19 7- 20 7- 21 7- 22 7-23 7-24 7-25 7-2
6
7-2
7
9-2
1
9-2
5 9 1
9-2 0-0 0-0
5
0-0
9
0-1
3
0-1
7
0-2
1
9-0 9-0 9-0 9-0 9-0 9-0 9-0 9-0 9-0 9-0 9-0 9-0 9-1 9-1 9-1 9-1 9-1 9-1
201 201 201 201 201 201 201 201 201 201 201 201 201 201 201 201 201 201
MCI Control Dementia Control
(a) House A (b) House B
Figure 3: The complexity of localisation.
50
60
70
RSSI
80
90
100
110
0:00 0:30 1:00 1:30 2:00 2:30 3:00 3:30 4:00 0 :00 0 :30 1:00 1:30 2:00 2:30 3:00 3:30 4:00
25 0 25 0 25 0 25 0 25 0 25 0 25 0 25 0 25 0 25 0 25 0 25 0 25 0 25 0 25 0 25 0 25 0 25 0
MCI Control
kitchen_1 bedroom_1 living_room
Figure 4: Visualisation of sleep disturbance for House A, and PwD (left), partner (right)
00:00:00-06:00:00 0.773 06:00:01-08:00:00 0.773
0.015 0.06
0.010 0.04
0.005 0.02
08:00:01-10:00:00 10:00:01-14:00:00
0.15 0.667 0.06 0.550
0.10 0.04
0.05 0.02
0.00 0.00
14:00:01-18:00:00 18:00:01-20:00:00
0.08 0.773 0.100 0.870
0.06 0.075
0.04 0.050
0.02 0.025
0.00 0.000
20:00:01-22:00:00 22:00:01-23:59:59
0.03 0.864 0.020 0.682
0.02 0.015
0.010
0.01
0.005
20 28 09 14 19 20 28 09 14 19
09- 09- 10- 10- 10- 09- 09- 10- 10- 10-
Dementia Control
Figure 5: House B - Comparing complexities of localisation.
Our data collection is ongoing, and as it grows we will identify finer, compared to that of the other participant during 0-6 am. Although
more informative signals of disease state. Time lags are likely to there is not a significant difference between activity levels in bedroom
become more relevant with finer and more targeted analysis that may 1 of the two inhabitants during 0-6 am in Fig. 2a, it is because that the
be motivated by changes in behaviour as the disease progresses. person with MCI got up earlier than 6 am, especially on July 23rd,
24th and 25th, and started walking in the house while the partner was
4.2 Wandering sleeping, which may suggest the increased sleep disturbance in the
person with cognitive disorder from another angle.
The approach for modelling wandering is motivated by its antonym: For reference, we also include a segment of the RSSI data in Fig. 4,
non-wandering is typified by low activity levels and persistent loca- from House A, for both participants, from the early hours of the day.
tion predictions. Thus we jointly compare the activity levels and the It highlights the difference in sleeping quality, as observed through
location complexity of residents in this section. Specifically, in Fig. RSSI. The step changes are a result of rotations during sleep. It also
2 we track the total activity per room, as well as the total activity of shows how different sleeping positions can lead to missing values in
the day, over the span of the data, and in Fig. 3 we track the complex- certain gateways, e.g. kitchen_1 at 02:00 and at 03:15.
ity of the localisation predictions, derived from the Lempel-Ziv [21]
complexity measure. It is clear in Fig. 2b that the activity levels of the
patient are higher than that of the other participant in this house, and 5 Conclusion and Future Work
Fig. 3b gives similar results, although in Fig. 2a and Fig. 3a it is less This paper presents our preliminary results and shows the poten-
obvious, for which there might be several reasons. One explanation tial of detecting early symptoms of cognitive disorders automatically
could be that the data we have so far for house A spans only a short by utilizing data acquired from patients’ daily activities and signal
time period (less than 10 days). A second cause might be that the processing methods equipped with machine learning techniques. We
patient in house A only has MCI, whose daily life only gets affected build a machine learning pipeline that considers the difference in
to a mild degree compared to the participant with dementia in house training and testing distributions using MMD. We account for the se-
B. quential nature of data using a CRF, and use our domain knowledge
Additionally, Fig. 5 shows in detail the complexity of localisation to enhance our training data with data coming from the participants.
predictions in different time intervals over the experiment span in One of the limitations of this study is that our localisation preci-
house B. On every subplot, the number at the top-right is the average sion can only reach room-level for now due to the configuration and
number of times the patient has a higher complexity than the partner. the deployment of our sensor system. Finer-grained localisation will
In this analysis, we only include house B, as the span is longer. In expose behavioural reactions to furniture placements across rooms.
the case of house A these averages are not so much in favour of the Potential directions for improvements include taking into consid-
patient - 5/8 segment as compared to 8/8 for house B. eration environmental sensors, such as Passive Infrared and Video,
through machine learning techniques such as multi-modal data fusion
4.3 Sleep Disturbance [9]. Moreover, we will enhance our training data using clinical data
through our domain knowledge of activities of daily living.
In both houses, we observe that during the early hours of the day,
The paper as it stands establishes baseline activity, sleep and co-
0-6 am, the participants have a higher activity levels, while the con-
localisation dependence that will be used after data collection com-
trol/carers have a more steady, and at most times, lower activity lev-
pleted. That we see signals here is promising for our future data
els. This might be an indication of the patients having a more uneasy
collection, analysis and research.
sleep. As a number of studies suggest [25, 15], people have MCI or
dementia will suffer from different levels of sleep disturbance. And
a community-based study [16] indicates that nearly one-third of their Acknowledgements
subjects were reported as having been inappropriately active at night.
This research was funded by CUBOID (UK MRC Momentum grant
As can be seen from Fig. 2b, the line representing the activity
MC/PC/16029) and the SPHERE IRC (grant EP/K031910/1).
levels in bedroom 2 of the participant with dementia is quite unstable
REFERENCES in 2018 IEEE Wireless Communications and Networking Conference
Workshops (WCNCW), pp. 302–307, (2018).
[20] Michal Kozlowski, Raul Santos-Rodriguez, and Robert J. Piechocki,
‘Sensor modalities and fusion for robust indoor localisation’, EAI En-
[1] Gregory D Abowd, Christopher G Atkeson, Aaron F Bobick, Irfan A dorsed Transactions on Ambient Systems, 6(18), (12 2019).
Essa, Blair MacIntyre, Elizabeth D Mynatt, and Thad E Starner, ‘Living [21] Abraham Lempel and Jacob Ziv, ‘On the complexity of finite sequences’,
laboratories: the future computing environments group at the georgia in- IEEE Transactions on information theory, 22(1), 75–81, (1976).
stitute of technology’, in CHI’00 Extended Abstracts on Human Factors [22] Ting-Ying Li, Chao-Lin Wu, Yi-Wei Chien, Li-Chen Fu, Chi-Chun
in Computing Systems, pp. 215–216, (2000). Chou, Chun-Chen Chou, and I-An Chen, ‘A supporting system for
[2] Ahmad Akl, Jasper Snoek, and Alex Mihailidis, ‘Unobtrusive detection quick dementia screening using pir motion sensor in smart home’, in
of mild cognitive impairment in older adults through home monitoring’, 2017 IEEE International Conference on Systems, Man, and Cybernetics
IEEE journal of biomedical and health informatics, 21(2), 339–348, (SMC), pp. 1369–1374. IEEE, (2017).
(2015). [23] Mary L Lilly, Melinda Hermanns, Jan Beckstrand, Adam Booth, and
[3] Ahmad Akl, Babak Taati, and Alex Mihailidis, ‘Autonomous unob- Martin R Farlow, ‘Illumination of shadowing behavior in individu-
trusive detection of mild cognitive impairment in older adults’, IEEE als with alzheimer’s disease: Proximity-seeking? as life becomes the
transactions on biomedical engineering, 62(5), 1383–1394, (2015). “strange situation”’, The Internet Journal of Neurology, 13(2), (2011).
[4] Barry Brumitt, Brian Meyers, John Krumm, Amanda Kern, and Steven [24] Constantine G Lyketsos, Maria C Carrillo, J Michael Ryan, Ara S
Shafer, ‘Easyliving: Technologies for intelligent environments’, in In- Khachaturian, Paula Trzepacz, Joan Amatniek, Jesse Cedarbaum,
ternational Symposium on Handheld and Ubiquitous Computing, pp. Robert Brashear, and David S Miller. Neuropsychiatric symptoms in
12–29. Springer, (2000). alzheimer’s disease, 2011.
[5] Dallan Byrne, Michal Kozlowski, Raul Santos-Rodriguez, Robert [25] Bryce A Mander, Joseph R Winer, William J Jagust, and Matthew P
Piechocki, and Ian Craddock, ‘Residential wearable rssi and accelerom- Walker, ‘Sleep: a novel mechanistic pathway, biomarker, and treatment
eter measurements with detailed location annotations’, Scientific Data, target in the pathology of alzheimer’s disease?’, Trends in neurosciences,
5, (August 2018). 39(8), 552–566, (2016).
[6] Diane J Cook, ‘Health monitoring and assistance to support aging in [26] Guy M McKhann, David S Knopman, Howard Chertkow, Bradley T
place.’, J. UCS, 12(1), 15–29, (2006). Hyman, Clifford R Jack Jr, Claudia H Kawas, William E Klunk, Wal-
[7] Ana Lígia Silva De Lima, Tim Hahn, Luc JW Evers, Nienke M De Vries, ter J Koroshetz, Jennifer J Manly, Richard Mayeux, et al., ‘The diag-
Eli Cohen, Michal Afek, Lauren Bataille, Margaret Daeschler, Kasper nosis of dementia due to alzheimer’s disease: recommendations from
Claes, Babak Boroojerdi, et al., ‘Feasibility of large-scale deployment the national institute on aging-alzheimer’s association workgroups on
of multiple wearable sensors in parkinson’s disease’, PLoS One, 12(12), diagnostic guidelines for alzheimer’s disease’, Alzheimer’s & dementia,
(2017). 7(3), 263–269, (2011).
[8] Tom Diethe, Mike Holmes, Meelis Kull, Miquel Perello Nieto, Kacper [27] Alex J Mitchell and Mojtaba Shiri-Feshki, ‘Rate of progression of mild
Sokol, Hao Song, Emma Tonkin, Niall Twomey, and Peter Flach, ‘Re- cognitive impairment to dementia–meta-analysis of 41 robust incep-
leasing ehealth analytics into the wild: Lessons learnt from the sphere tion cohort studies’, Acta Psychiatrica Scandinavica, 119(4), 252–265,
project’, in Proceedings of the 24th ACM SIGKDD International Con- (2009).
ference on Knowledge Discovery & Data Mining, pp. 243–252, (2018). [28] Alexander M. Rush. Torch-struct: Deep structured prediction library,
[9] Tom Diethe, Niall Twomey, Meelis Kull, Peter Flach, and Ian Craddock, 2020.
‘Probabilistic sensor fusion for ambient assisted living’, arXiv preprint [29] Volker Schwarz, Alex Huber, and Michael Tuchler, ‘Accuracy of a com-
arXiv:1702.01209, (2017). mercial uwb 3d location/tracking system and its impact on lt application
[10] Matthew D’Souza, Montserrat Ros, and Mohanraj Karunanithi, ‘An scenarios’, in 2005 IEEE International Conference on Ultra-Wideband,
indoor localisation and motion monitoring system to determine be- pp. 599–603. IEEE, (2005).
havioural activity in dementia afflicted patients in aged care’, (2012). [30] Emma Stack, Veena Agarwal, Rachel King, Malcolm Burnett, Fate-
[11] Atis Elsts, Tilo Burghardt, Dallan Byrne, Massimo Camplani, Dima meh Tahavori, Balazs Janko, William Harwin, Ann Ashburn, and Dorit
Damen, Xenofon Fafoutis, Sion Hannuna, William Harwin, Michael Kunkel, ‘Identifying balance impairments in people with parkinson’s
Holmes, Balazs Janko, et al., ‘A guide to the sphere 100 homes study disease using video and wearable sensors’, Gait & posture, 62, 321–
dataset’, arXiv preprint arXiv:1805.11907, (2018). 326, (2018).
[12] Xenofon Fafoutis, Antonis Vafeas, Balazs Janko, R Simon Sherratt, [31] Charles Sutton, Andrew McCallum, et al., ‘An introduction to condi-
James Pope, Atis Elsts, Evangelos Mellios, Geoffrey Hilton, George tional random fields’, Foundations and Trends® in Machine Learning,
Oikonomou, Robert Piechocki, et al., ‘Designing wearable sensing plat- 4(4), 267–373, (2012).
forms for healthcare in a residential environment’, EAI Endorsed Trans- [32] Niall Twomey, Tom Diethe, Ian Craddock, and Peter Flach, ‘Unsuper-
actions on Pervasive Health and Technology, 3(12), (2017). vised learning of sensor topologies for improving activity recognition
[13] Michael Grundman, Ronald C Petersen, Steven H Ferris, Ronald G in smart environments’, Neurocomputing, 234, 93–106, (2017).
Thomas, Paul S Aisen, David A Bennett, Norman L Foster, Clifford R [33] Niall Twomey, Tom Diethe, Xenofon Fafoutis, Atis Elsts, Ryan Mc-
Jack Jr, Douglas R Galasko, Rachelle Doody, et al., ‘Mild cognitive Conville, Peter Flach, and Ian Craddock, ‘A comprehensive study of
impairment can be distinguished from alzheimer disease and normal activity recognition using accelerometers’, in Informatics, volume 5,
aging for clinical trials’, Archives of neurology, 61(1), 59–66, (2004). p. 27. Multidisciplinary Digital Publishing Institute, (2018).
[14] Sumi Helal, William Mann, Hicham El-Zabadani, Jeffrey King, Youssef [34] Antonis T Vafeas, Xenofon Fafoutis, Atis Elsts, Ian J Craddock, Md Is-
Kaddoura, and Erwin Jansen, ‘The gator tech smart house: A pro- rafil Biswas, Robert J Piechocki, and George Oikonomou, ‘Wearable
grammable pervasive space’, Computer, 38(3), 50–60, (2005). devices for digital health: The sphere wearable 3’, in Embedded Wire-
[15] Eva Hita-Yañez, Mercedes Atienza, Eulogio Gil-Neciga, and Jose L Can- less Systems and Networks (EWSN): On-Body Sensor Networks (OBSN
tero, ‘Disturbed sleep patterns in elders with mild cognitive impairment: 2020), (2020).
The role of memory decline and apoe 𝜀 4 genotype’, Current Alzheimer [35] Przemyslaw Woznowski, Alison Burrows, Tom Diethe, Xenofon
Research, 9(3), 290–297, (2012). Fafoutis, Jake Hall, Sion Hannuna, Massimo Camplani, Niall Twomey,
[16] RA Hope and Christopher G Fairburn, ‘The nature of wandering in Michal Kozlowski, Bo Tan, et al., ‘Sphere: A sensor platform for health-
dementia: A community-based study’, International journal of geriatric care in a residential environment’, in Designing, Developing, and Facil-
psychiatry, 5(4), 239–245, (1990). itating Smart Cities, 315–333, Springer, (2017).
[17] Jiayuan Huang, Arthur Gretton, Karsten Borgwardt, Bernhard [36] Ni Zhu, Tom Diethe, Massimo Camplani, Lili Tao, Alison Burrows,
Schölkopf, and Alex J Smola, ‘Correcting sample selection bias by Niall Twomey, Dritan Kaleshi, Majid Mirmehdi, Peter Flach, and Ian
unlabeled data’, in Advances in neural information processing systems, Craddock, ‘Bridging e-health and the internet of things: The sphere
pp. 601–608, (2007). project’, IEEE Intelligent Systems, 30(4), 39–46, (2015).
[18] William D Kearns, Donna Algase, D Helen Moore, and Sadia Ahmed,
‘Ultra wideband radio: A novel method for measuring wandering in
persons with dementia’, Gerontechnology, 7(1), 48, (2008).
[19] Michal Kozlowski, Dallan Byrne, Raul Santos-Rodriguez, and Robert
Piechocki, ‘Data fusion for robust indoor localisation in digital health’,