INDOOR LOCALISATION FOR DETECTING MEDICA-TION USE IN PARKINSON'S DISEASE

Abstract

Parkinson's disease (PD) is a slowly progressive debilitating neurodegenerative disease which is prominently characterised by motor symptoms. Indoor localisation, including its in-home mobility features, could provide a digital biomarker that can be used to quantify how mobility changes as this disease progresses. To improve the effectiveness of current methods for indoor localisation, a transformer-based approach utilising multiple modalities, Received Signal Strength Indicator (RSSI) and accelerometer data from wearable devices, which provide complementary views of movement, is proposed. To properly evaluate our proposed method, we use a free-living dataset where the movements and mobility are greatly varied and unstructured as expected in real-world conditions. 12 pairs of people (one with PD, and the other a control participant) lived for five days in a smart home with various sensors. Our evaluation on such a dataset, which includes subjects with and without PD, demonstrates that our proposed network outperforms the current state-of-the-art in indoor localisation. We also show how the accurate room-level localisation predictions can be transformed into in-home mobility features (i.e. room-to-room transition duration) which can be used to effectively classify whether the PD participant is taking their medications or withholding them (increasing their symptoms).

1. INTRODUCTION

Parkinson's disease (PD) is a neurodegenerative disorder that affects around six million people worldwide. It is a chronic disease with four main symptoms: tremor, bradykinesia (slowness of movement), rigidity, and postural instability Jankovic (2008) . Although PD is a slowly progressive disease, the symptoms can fluctuate hourly depending on medication intake timing, stress and other factors. These fluctuations make it challenging to capture the slow symptom progression in an individual (needed for example to measure how a new therapy changes disease progression) accurately over time using infrequent clinic or lab-based "snapshot" clinician-patient interactions. Also, to help a clinician understand the impact of PD on a patient, continuous monitoring is needed to give an accurate evaluation of how severe the symptoms and their fluctuations are for that individual. One method to determine whether sensor based monitoring can detect symptom fluctuations in PD is by evaluating the patients in an "OFF" medication state (i.e. when they have not taken their medication) since, when they withhold medications, patients tend to experience a worsening of symptoms. For example, when "OFF" medications, motor symptoms can become more severe which may hinder the subject's gait and movement around their own home. As a result, they may typically need more time to transition between rooms. Indoor localisation can be used to measure transitions between rooms in a home. Knowing how slow or fast a person transitions between rooms may be able to predict whether a person is in an "OFF" medication state or not and, longer-term, interpret their symptoms' severity (e.g. de novo freezing of gait in a hallway could indicate symptom progression) Bachlin et al. (2009) . Localisation can also add context in the measurement of other behaviours such as urinary function monitoring (e.g., how many times someone visits the bathroom overnight) which can impact the quality of life with PD He et al. (2016) . To perform indoor localisation in home environments, IoT-based platforms with multimodal sensors combined with machine learning can be used to provide an unobtrusive and continuous localisation. Typically, a wearable inertial measurement unit (IMU) is used to produce a received signal strength indicator (RSSI) at points throughout a home, which can be used to create radio-map features for room-level localisation. To provide more accurate localisation, accelerometer data from the same wearable can also be used as it provides a means to distinguish different activities (e.g., walking vs standing). As some activities are tied to particular rooms (e.g. stirring a pan on the hob must be in a kitchen), accelerometer data may enrich RSSI in differentiating adjacent rooms, which RSSI alone may struggle with Jovan et al. (2022) . If accelerometer data are to provide extra features for separating adjacent rooms, greater consideration must be given to data generalisation across different PD patients. As PD is a heterogeneous disease, the severity of symptoms may vary from one patient to another Greenland et al. (2019) . These severe symptoms, such as tremor, may affect the generalisation of accelerometer data, especially those worn on the patient's wrists, which is a common and well accepted placement location. Naively combining the accelerometer data with the RSSI may impair the performance of indoor localisation due to differing levels of tremor manifesting in the acceleration signal. In this work, we make two main contributions. (1) We describe the utilisation of RSSI enriched by the accelerometer data to perform room-level localisation. Similar to Jovan et al. ( 2022), our proposed network intelligently chooses accelerometer features which may improve the RSSI performance in performing indoor localisation. To properly evaluate our proposed method, we use a free-living dataset where the movements and mobility are greatly varied and unstructured as expected in real world conditions. Our evaluation on such a dataset, which includes subjects with and without PD, demonstrates that our proposed network outperforms the current state-of-the-art DCMN (Jovan et al. ( 2022)) in all categories. (2) We also show how the accurate room-level localisation predictions can be transformed into in-home mobility features (i.e. room-to-room transition duration) which can be used to effectively classify the "OFF" or "ON" medication state of a PD patient. 2020) match video sequences of silhouettes to accelerations from wearable sensors for a person re-indentification in a home environment. Notably, all of this research uses vision as their main data source; although powerful and rich in features, vision sensors typically raise privacy challenges in home settings and are typically absent in key rooms within the home for this reason.

2. RELATED WORK

RSSI data produced from wearable devices is one such mechanism with less privacy concerns. In indoor localisation, fingerprinting using RSSI is the typical technique used to estimate the location of wearable devices by exploiting this signal that can be measured in the environment. RSSI signals are not stable, they fluctuate randomly due to shadowing, fading and multi-path effects. However, many techniques have been proposed in recent years to tackle these fluctuations, and, indirectly, improve the localisation accuracy. Some of the work, Zhang et al. ( 2016 



Early work in machine learning for monitoring a long-term neurodegenerative disease like PD started with a simple PD classification Fraiwan et al. (2016) or easy-to-distinguish symptoms identification Arora et al. (2015); Fisher et al. (2016) using one single sensor modality such as vision sensors for simplicity Li et al. (2018). As the research progressed, multiple sensors have been considered for performance improvement in identifying PD symptoms. For example, Heidarivincheh et al. (2021) utilised vision and accelerometer sensors available in a smart home to classify PD from non-PD. Masullo et al. (

), utilise deep neural network (DNN) to generate coarse positioning estimates from RSSI signals, which are then refined by a hidden Markov model (HMM) to produce a final estimate location. Other works, Ibrahim et al. (2018), try to utilise a time-series of RSSI data and exploit the temporal connections within each access point to estimate room-level position. A CNN is used to build localisation models to further leverage the temporal dependencies across time-series readings. It has been suggested that we cannot rely on RSSI alone for indoor localisation in home environments for PD subjects due to shadowing rooms with tight separation Pandey et al. (2021). Jovan et al. (2022) et al. combine RSSI signals (for location estimate) and accelerometer data (for body movement estimate) to produce a more accurate location estimate by utilising a wider range of features to differentiate adjacent rooms. Their proposed network has shown an improvement in tracking a person with PD in a hallway that, in turn, can be used to track their transition time between rooms. Our work is inspired by Jovan et al. (2022) with two main improvements. While Jovan et al. capture

