A preliminary study of movement intensity during a Go/No-Go task and its association with ADHD outcomes and symptom severity

This was a two-phase cross-sectional study. The first phase included both an ADHD and a control group to assess the discriminating capabilities of movement intensity measures extracted from data collected by a Microsoft Kinect System. The second phase of the study included only a subset of the ADHD group and was designed to explore associations between movement intensity measures and ADHD symptomatology.

Subjects were girls and boys aged 6–12 years living in Beijing city. Children in the ADHD group were selected to participate if they met diagnostic criteria for any presentation of ADHD (inattentive, hyperactive-impulsive, or combined) according to DSM-5 criteria [2] or who were considered to be subthreshold for ADHD, defined as one symptom short of meeting diagnostic criteria. Children with ADHD were excluded if any other co-morbid psychiatric condition (e.g., anxiety disorder, depression) was present. A subset of ADHD cases (N = 14) were recruited from a randomized, wait-list controlled, multi-site study entitled, the “Integrated Brain, Body and Social (IBBS) Intervention for Attention-Deficit/Hyperactivity Disorder” (ClinicalTrials.gov Identifier: NCT01542528; IBBS study) [24] whereas the rest of the ADHD participants (N = 16) were outpatients from a psychiatric hospital serving Beijing City. Children in the control group were matched to children in the ADHD group according to age and gender and were recruited from a local elementary school.

A total of 60 children were enrolled in phase I of the study. Thirty children were in the ADHD group and 30 children were in the control group. All participants were of Han ancestry and each group consisted of 28 boys and 2 girls. The mean age for both groups was 8.95 years (SD = 1.88). The ADHD group consisted of 19 children with ADHD combined type, 5 with inattentive type, 4 with hyperactive-impulsive type, 1 with subthreshold combined type, and 1 with subthreshold inattentive type based on in-person clinical evaluations. One child in the ADHD group had discontinued treatment with methylphenidate (10 mg) due to side effects for 6 months prior to participation in the study.

In phase 2, a total of 14 children from the IBBS study with ADHD or subthreshold for ADHD (9 ADHD combined type, 1 inattentive type, 2 hyperactive-impulsive type, 1 subthreshold combined type, and 1 subthreshold inattentive type) participated. The mean age of the sample was 7.32 years (SD = 1.02). Except for the one child referred to above, all participants were medication naive. Considering ADHD symptom severity ratings were completed only for participants from the IBBS study as part of the assessment protocol and not for those participants recruited from the outpatient clinic, the sample in phase II of the study was limited to just the IBBS study participants.

Body movements during a Go/No-Go task were monitored and recorded by a Microsoft Kinect infrared motion sensing camera. This camera was placed 150 cm from the child at a 45^° angle from the line between the child and a laptop computer that was used to present the Go/No-Go task (Fig. 1). To ensure the quality of sampling, children were restricted to standing in a circle with a radius of approximately 25 cm [33]. The Kinect camera is a horizontal bar connected to a small base with a motorized pivot and consists of a Red–Green–Blue camera and depth sensor. The camera has a pixel resolution of 640 × 480 and a frame rate of 30 frames per second (FPS). The image depth sensor contains a monochrome complementary metal oxide semiconductor (CMOS) and an infrared projector, which emits multiple infrared rays to form a close-spaced light spot matrix in order to determine its distances from multiple reference points of a participant’s silhouette. The data from this depth sensor were then pre-processed to create a 3-dimensional bitmap that allowed for the monitoring of pixels by comparing temporally adjacent frames to detect movement and extract measures of movement intensity [34].

Both phases of this study were approved by an ethics review board (Scientific Research Ethics Committee of the Institute of Psychology, Chinese Academy of Sciences Beijing, P.R. China). Informed consent was obtained from parents and all child participants gave informed assent prior to initiating any study procedures. For those ADHD participants recruited from the IBBS study, best-estimate DSM-5 diagnoses were assigned by two experienced psychiatrists following a clinical interview with participants’ parents using the Chinese version of the Kiddie Schedule for Affective Disorders and Schizophrenia—Present and Lifetime Version (K-SADS-PL, [35, 36]). ADHD symptom severity ratings were also provided by two expert clinicians as part of the IBBS assessment battery. Once study eligibility was confirmed, participants completed a Go/No-Go Task while the Microsoft Kinect System monitored and recorded their bodily movements. All study procedures for this subset of ADHD participants including the collection of movement data occurred during the IBBS screening visit. The collection of movement data for the remaining ADHD participants took place after their diagnoses were confirmed at the outpatient psychiatric hospital. Diagnoses were made by two experienced psychiatric clinicians based on a clinical interview with the children’s parents, parents’ ratings on a measure assessing their children’s emotions and behavior (i.e., Achenbach Child Behavior Checklist [16]) and an attention task (i.e., Cross-out task [37]). Children from the control group participated in study procedures during one visit to their school by the research team after written consent/assent was given. To confirm the typical development of participating children, their clinical files containing classroom behavior history and routine mental health sessions were reviewed by the school psychologist. A brief screening interview of DSM-5 diagnoses was also done independently by an experienced psychiatrist at the local hospital to confirm their “healthy control” designation. All the movement data were collected in private rooms with the curtains drawn to limit distractions and control the environment’s light so that the children could see the monitor screen clearly.

This study used bitmap source data of participants’ silhouettes including depth information from the Microsoft Kinect system. The raw silhouette data can be quite unstable and inconsistencies can be observed when viewing the frames in sequence, as noise fragments can be observed bursting across the silhouette even when participants are standing completely still. The noise level of Microsoft Kinect’s infrared sensor has shown to be correlated with the distance between the sensor and target [38] so by keeping this distance constant, one source of noise was minimized. To further account for the remaining noise, a denoise procedure was used to extract the movement intensity measures. First, a baseline assessment of movement was conducted by asking participants to stand still for 15 s. As the average noise level across all 60 participants was 25 pixels (SD = 3.1) when standing still, a scan-line algorithm was used to remove regions of noise smaller than 25 pixels from each participant’s recording. The Kinect data was then preprocessed by comparing two temporally adjacent bitmaps of the silhouette pixel-by-pixel, to determine if there was a change between the two frames (see Additional file 1: Figure S1). Within a given time interval, if a particular pixel had different spatial coordinate values than the previous frame, the program was instructed to mark it as a moved pixel. This yielded a movement intensity value across two adjacent frames where a greater number of moved pixels was indicative of greater intensity in the movement between two frames.

Considering the total pixel count that represented a child’s body was continually changing due to movement, it was necessary to transform the moved pixel count into a converted score by dividing the total moved pixel count by the total mass of the child’s body (i.e., number of pixels representing the child’s silhouette in the current fame).

This converted value of movement intensity was recorded for each frame. As this value was time-locked, it represented a time domain signal to which a Fourier transformation was applied to produce a movement intensity distribution (MID). Since the Kinect camera has a sampling rate of 30 Hz, the frequency domain resolution was expected to be half this sampling rate, resulting in a 0–15 Hz range. The MID data was then subdivided into 15 non-overlapping 1 Hz frequency bands (FB). Thus, the following measures were calculated from the data captured by the Microsoft Kinect System: a composite measure of total movement intensity (TMI) and a movement intensity distribution (MID) across 15 frequency bands (the FB measures).

Children in the control group had significantly better performance across all six performance measures on the Go/No-Go task as compared to the ADHD group (Table 1). The ADHD group displayed more movement than the control group, as group comparisons were all statistically significant (p < 0.05) for the TMI and FB measures even after applying the FDR adjustment. The AUC was 0.904 for the TMI measure and between 0.867 and 0.932 for the 15 FB measures indicating that these measures of movement intensity had adequate to good precision with regard to accurately classifying children with and without ADHD (Fig. 2). Overall, 29 of 30 children with ADHD were discriminated from 25 of 30 normal controls with a sensitivity of 0.967 and specificity of 0.833, as calculated using the TMI measure. The ROC-AUC analysis for Go/No-Go task measures revealed AUC values between 0.69 and 0.93 with reaction time variability having the best discriminability: AUC of 0.93, sensitivity of 0.967, and specificity of 0.867. Only commission errors on the Go/No-Go task were significantly correlated with the TMI measure (r = 0.28, p = 0.03).

After applying the FDR adjustment, 12 out of 15 frequency bands were correlated with the CGI-S scores and 10 out of 15 bands were correlated with the ADHD-RS total scores, 10 out of 15 bands were correlated with the ADHD-RS hyperactivity subscale and 7 out of 15 bands were correlated with the ADHD-RS inattentive subscale. The 10–11 and 12–13 Hz frequency bands had the strongest correlations with the ADHD-RS (total and subscales) and CGI-S scores (Table 2). The TMI measure was not correlated with the ADHD-RS total scores or either the hyperactivity or inattentive subscale scores, but it was significantly correlated with the CGI-S scores (r = 0.61, p = 0.021). There were no significant correlations between any of the Go/No-Go performance measures and ADHD symptom severity measures [ADHD-RS (total and subscales) and CGI-S].