Freezing of gait (FoG) is a widespread Parkinson's disease (PD) symptom. It is a disabling, complex, and highly variable clinical phenomenon characterized by brief episodes of inability to step or very short steps (Nutt et al., 2011). This condition usually leads to falls (Paul et al., 2013), falls-related injury, and ultimately reduced independence in the activities of daily living (Gilat et al., 2018). FoG is often described as the sensation of having the feet glued to the floor while the body center of mass continues to move forward. There is a growing interest in non-pharmacological interventions to manage FoG, and reliable tools are required to determine the severity of this symptom. Wearable sensors are emerging as new tools to obtain information about FoG objectively. A typical wearable sensor consists of a microelectromechanical system with a gyroscope (angular position rate sensor), an accelerometer (acceleration sensor), and a magnetometer (orientation sensor) with one, two, or three axes at each sensor. Several automatic freeze-detection algorithms based on some or all signals of one or multiple wearable sensors have been developed (Silva De Lima et al., 2017; Pardoel et al., 2019). These methods vary in complexity, from frequency domain analysis (Moore et al., 2008, 2013; Delval et al., 2010; Capecci et al., 2016), wavelet transforms (Punin et al., 2019) to multiple features and machine-learning techniques (Tripoliti et al., 2013; Ahlrichs et al., 2016).

A major limitation to developing robust algorithmic methods for FoG estimation is the lack of a large public data set presenting kinematic data recorded on PD patients. To the best of our knowledge, there are four open data sets¹ (Bachlin et al., 2010; Mazilu et al., 2013; Li, 2021; Pardoel et al., 2021) with some of these characteristics. For example, (a) The Daphnet (Bachlin et al., 2010) contains acceleration data for 10 PD patients in several laboratory walking tasks, and (b) CuPiD (Mazilu et al., 2013) included 35 motion and three physiological sensors in 18 PD patients. Both datasets were recorded during walking and studied FoG features of walking.

Our current work fills the gap in two critical points. First, none of these studies included data during the turning phase; this is an important limitation as studies have shown that FoG events are very frequent during the turning phase in PD patients (Mancini et al., 2017). Second, given the variability of the patients' conditions (severity of the PD, level of medication), precise control of the clinical and medication status is necessary. Since individuals remain ON medication for most of the day, the dataset should include measurements in this medication condition (the Daphnet presents data during OFF medication condition, and no information is provided in CuPiD).

Here we present an open data set of PD patients, which includes both (a) study patients' demography, (b) clinical conditions (PD severity, the number and duration of FoG episodes for each individual, clinical scales, and medication state during testing), and kinematics (video, acceleration, and angular velocity) during a turning-in-place task in individuals with PD in the ON medication state.

The data collection was performed in the School of Physical Education and Sport at the University of São Paulo, Brazil. The local ethics committee approved this study, and all patients signed a consent form before the data collection. As FoG is sporadic, the PD patients participated in three experimental sessions at an interval of 1 month between each session to increase the chances of occurrence. Since medication can influence the presence of the FoG and that most PD subjects are ON medication, we performed our measurement during this ON medication condition, meaning a stable dose of antiparkinsonian medication for at least 1 month, and they had taken dopaminergic medication 1 h before starting the sessions to ensure dose stabilization.

Participants spent a total of 1,611 s in FoG. On average, participants experienced 3.0 (SD = 2.9) FoG episodes, but 36% did not have FoG episodes during the turning task. The data is available at Figshare (https://doi.org/10.6084/m9.figshare.14984667) under the CC0 4.0 license. The data set contains three types of information: (a) metadata, (b) separate text files were generated for the IMU processed data for each session, and (c) the video of each session. In addition, the clinical characteristics of the patients are presented in Table 1.

We presented an open data set of acceleration and angular velocity data from wearable sensors placed on one leg, the video during the turning task, plus a file with metadata containing clinical information in individuals with PD. In addition, data were generated from the evaluation of 35 PD patients with FoG. Furthermore, the open data set includes the identification of FoG carried out by movement disorders specialists using videos for qualitative analysis.

Our results demonstrated a significant correlation between subjective (NFoG-Q) and objective (FoG-ratio, and total time of FoG during turning task) FoG measures, as observed previously (Mancini et al., 2017). This result is interesting for two reasons. First, NFoG-Q is related to a patient's perception of FoG duration and frequency that they had experienced during the past month; however, it is subjective and may represent a weak indication of FoG severity out of the stimulating clinic environment (Mancini et al., 2019). Second, despite the turning task being an objective lab-based method to measure freezing severity without considering the patient's self-perception, FoG episodes may disappear during the lab examination due to the patient paying extra attention to gait (Mancini et al., 2019). Even so, the finding that both FoG measures are associated confirms that the patients' judgment of their freezing severity was associated with FoG severity assessed by an objective lab-based method.

To the best of our knowledge, this is the first data set that includes, in addition to kinematic data, cognitive or psychiatric scales in patients with PD and FoG. These scales are extremely important when studying FoG. For example, anxiety was the strongest predictor of FOG development after 15 months of the initial diagnosis (Ehgoetz Martens et al., 2018). Thus, anxiety may act as a stressor and trigger FOG in daily activities, as FOG occurs at home (Mancini et al., 2019), and patients may have feelings of inability to perform any activities due to lack of confidence and fear of falling.

Previous FoG datasets with wearable sensors have been published elsewhere (Bachlin et al., 2010; Mazilu et al., 2013). Although these studies presented valuable information, the open data set we showed in this study is unique in the literature, given the medication status and a complete description of the clinical status of the participants. One exciting and promising future step for researchers could be developing new FoG detection algorithms that contain the patients' clinical and physiological information. Despite the patients being classified as FoG, 36% did not have FoG episodes during the turning task. This information could be used as a control in identifying patterns of FoG episodes. A limitation of our data set is the use of only one sensor; studies have shown that multi-sensor fusion systems can perform better than single-sensor systems (Bachlin et al., 2010; Li et al., 2020). Nevertheless, our data set will be interesting for the researcher to develop decoding algorithms with minimal hardware.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found below: 10.6084/m9.figshare.14984667.

The studies involving human participants were reviewed and approved by Federal University of ABC. The patients/participants provided their written informed consent to participate in this study. Written informed consent was obtained from the individual(s) for the publication of any potentially identifiable images or data included in this article.

CR: conceptualization, methodology, data curation, and writing—original draft. RM and JÁ: methodology and data curation. AL-P and DF: writing—original draft. CS-B and LT: writing—review and editing. SS: writing—original draft, writing—review, and editing. BM: project administration and funding acquisition. DC: conceptualization, methodology, data curation, writing—original draft, writing—review, and editing. All authors contributed to the article and approved the submitted version.

This study was supported by Leading House for the Latin American Region, the University of St. Gallen, Seed Money Grant, Switzerland.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.