A fuzzy logic system for the home assessment of freezing of gait in subjects with Parkinsons disease

Highlights

• A novel fuzzy logic algorithm for freezing of gait detection is presented.

• A smartphone app was developed to enhance usability and acceptability.

• High reliability in laboratory tests against clinicians observation.

• Home monitoring correlates significantly with laboratory clinical evaluation.

• A well known algorithm was applied on the same data for comparison.

Abstract

Gait dysfunctions are pathognomonic, progressive and, generally, continuous in Parkinson’s Disease (PD). The Freezing of Gait (FoG) is an episodic gait disorder involving up to 70% of people with PD, within 10 years of clinical onset, and associated with an increased risk for falls and immobility, which in turn, contributes to greater disability. Automatic and objective monitoring of FoG may help clinicians to understand and treat this phenomenon. In this work, a smartphone app for real-time FoG detection is presented and tested both in a laboratory setting and at patients’ home. The app implements a novel fuzzy logic algorithm that uses important spatio-temporal parameters of gait and is built according to clinical knowledge about FoG. The app includes a gait detection function and the evaluation of two important clinical statistics, i.e. FoG time and FoG number. The app FoG detection performance was assessed against clinicians evaluation and compared with the Moore–Bachlin FoG detection algorithm through ROC analysis, the calculation of confusion matrix, and FoG hit rate. The proposed algorithm achieved better results with respect to the Moore–Bachlin algorithm. Home reports were compared with respect to the FoG Questionnaire and laboratory reports; results indicated significant correlations for both FoG time and FoG number. The results confirm the reliability and accuracy of this app for FoG detection, supporting its wide use for diagnostic and therapeutic purposes.

Introduction

Parkinson’s Disease (PD) is the most common neurodegenerative disorder that, mainly, involves basal ganglia and dopaminergic neurons (Braak, Ghebremedhin, Rb, Bratzke, Del Tredici, 2004, Pringsheim, Jette, Frolkis, Steeves, 2014). This pathology causes a severe difficulty in motor planning, action and execution of non-attention demanding tasks (automaticity) (Bartels & Leenders, 2009). Typical motor disorder in PD are bradykinesia, tremor, rigidity, flexed posture, postural instability, festination and Freezing of Gait (FoG) (Jankovic, 2008). FoG may be regarded as a frequent and disabling condition, involving up to 70% PD people within 10 years of clinical onset (Tan, McGinley, Danoudis, Iansek, & Morris, 2011). It is a highly impairing and distressing motor block, defined as a ’brief, episodic absence or marked reduction of forward progression of the feet despite the intention to walk’ (Nutt et al., 2011). FoG impairs mobility, causes falls (Kerr, Worringham, Cole, Lacherez, Wood, Silburn, 2010, Latt, Lord, Morris, Fung, 2009), reduces quality of life and social participation (Moore, Peretz, & Giladi, 2007). The pathogenesis of FoG is still considered mysterious, given its episodic nature and heterogeneous manifestation (Nutt et al., 2011), and pharmacological treatments are of poor efficacy. For these reasons, FoG is an important clinical problem.

The study of kinematic and dynamic data collected in laboratory settings disclosed some FoG characteristics in frequency and time domain: high frequency ($3-8$ Hz) lower limbs trembling (Hausdorff, Balash, & Giladi, 2003); progressive reduction of Step Length (SL) joined to an increase in Step Cadence (SC) and a decrease in gait speed (Nieuwboer et al., 2001); reduced joint excursion of hip, knee and ankle (Nieuwboer et al., 2001); disordered temporal control of gait cycle (Nieuwboer, Chavret, Willems, & Desloovere, 2007). These clinical studies encouraged the development of systems able to detect and quantify FoG automatically. To this purpose, accelerometers and gyroscopes are suitable sensing units, since their low weight, dimensions and cost make them fitted for out-of-laboratory context. Nevertheless, home monitoring is poorly investigated in the literature (Silva de Lima et al., 2017) despite its crucial importance, since FoG is difficult to observe in laboratory. Some research groups (Kwon et al., 2014) highlight the importance of real-time FoG detection in order to give alerts for fall prevention or to deliver external stimuli, such as auditory cues or vibro-tactile signals that proved to be helpful to unfreeze gait and prevent FoG related falls (Baker, Rochester, & Nieuwboer, 2007). Despite the effort of researchers, the use of FoG detection systems in the clinical practice is still absent, likely due to the lack of user centered design, of meaningful and accessible reports about FoG for clinicians, and to heterogeneity of proposed systems (Silva de Lima et al., 2017).

The system proposed in this work tries to address the above mentioned issues by allowing the immediate delivery of reliable and meaningful data to clinicians, increasing usability and acceptability, and monitoring patients at their homes. In details, the most important contributions of the proposed FoG monitoring system are:

• a novel fuzzy logic FoG detection algorithm exploiting two spatio-temporal gait parameters mostly related with FoG, i.e. SC and SL, joined with other frequency domain features;

• real-time tests of the proposed algorithm implemented on a smartphone, used for both sensing and processing functions, in a sample of patients with PD;

• the validation of the FoG monitoring system against clinicians’ observation in the home environment as well as in the laboratory, and the comparison with the Moore-Bachlin (MB) algorithm (Bachlin, Plotnik, Roggen, Maidan, Hausdorff, Giladi, Troster, 2010, Moore, MacDougall, Ondo, 2008), which is the most used FoG detection algorithm.

Fuzzy logic leads to a model that clinicians can easily understand and interpret through its linguistic rules, providing important insights into this motor disorder and the opportunity to better optimize treatments. Fuzzy logic FoG detection was firstly explored by this research group to overcome some limitations introduced by crisp or threshold-based rules: a preliminary fuzzy inference system was tested offline on previously collected data (Pepa, Ciabattoni, Verdini, Capecci, & Ceravolo, 2014), however this preliminary attempt used only some frequency features. In this paper, the fuzzy logic algorithm is different and possibly improved by:

• adding significant spatio-temporal gait parameters to the input variables;

• introducing fuzzy set customization on the specific user;

• changing the fuzzy inference system architecture and rules.

Furthermore, the proposed monitoring system comprises the gait detection function and the evaluation of two important clinical statistics, i.e. FoG time (Tfog) and FoG number (Nfog), which were introduced in a previous preliminary study (Pepa, Capecci, Ciabattoni, Spalazzi, & Ceravolo, 2017a). These functions can enhance the delivery of meaningful reports for clinicians, especially during home monitoring where the system may acquire a lot of data not related to gait. This hypothesis may find confirmation through the laboratory and home experiments described here.

The paper is structured as follows. Section 2 reports the related work. The proposed fuzzy algorithm along with the overall architecture of the mobile app are presented in Section 3. The experimental results are reported in Section 4 and discussed in Section 5. Finally, Section 6 contains some concluding remarks.