Seventeen patients who had been diagnosed with early-stage AD (Age 64.9 ± 6.5 y, 7 male, 10 female) and 17 age- and gender-matched cognitive normal (CN, Age 64.9 ± 5.8 y, 7 male, 10 female) adults participated in the experiment. All subjects were right-handed with normal or corrected-to-normal vision. The handedness of each subject was based on their self-report followed by assessment of Edinburgh Handedness Inventory. The AD patients were recruited from the Department of Neurology at Qilu Hospital of Shandong Province, China. They were diagnosed as early stage of AD according to the criteria of National Institute of Neurological and Communicative Diseases and Stroke/Alzheimer's Disease and Related Disorders Association by professional therapists. The diagnosis and staging were based on comprehensive judgement according to neuropsychological tests, braining imaging, and amyloid-beta and tau in cerebrospinal fluid [2, 25]. Neuropsychological tests including the Mini–Mental State Examination (MMSE), the Montreal Cognitive Assessment (MoCA), the Hamilton Anxiety Scale (HAMA) and the Hamilton Depression Scale (HAMD) were performed on each AD patient. Inclusion criteria were: (1) over 50 years old; (2) clear mental state; and (3) ability to understand the instructions. Exclusion criteria were: (1) late-stage AD; (2) sever stroke; (3) Parkinson's disease; (4) history of upper-limb fractures or upper-limb diseases including but not limited to scapulohumeral periarthritis, scapular soreness, ulnar tunnel syndrome, radial tunnel syndrome, carpal tunnel syndrome, finger fractures, tenosynovitis, elbow ankylosis or peripheral neuropathies. Each subject was fully informed the purposes of this study and given informed consent prior to the experiment. The experimental procedures were approved by the Institutional Review Board of Shandong University (KYLL-2020(KS)-340) and were in accordance with the Declaration of Helsinki.

Retro-reflective markers were affixed to the dorsal surface of the right hand of each subject. The markers included nail marker-clusters on distal segments of the thumb and index finger [26, 27], hand marker-cluster along the second metacarpal, and single marker proximal to wrist. An optical three-dimensional (3D) motion capture system (OptiTrack™, USA) was used to track the position of the markers. A reflective marker installed on a base (Fig. 1a) was used as the grasping target for kinematic task. A custom-made apparatus installed with two six-component force/torque transducers (Nano 17, ATI Industrial Automation, Inc., Apex, NC) was used as the grasping target for the kinetic task (Fig. 1f). The transducers were mounted on the apparatus by precisely positioning that the x-axis and y-axis were along the vertical and horizontal directions in the contact surface of each transducer, and the z-axis was in the perpendicular direction to the contact surface (Fig. 1f). The grip surfaces with a span of 50 mm were covered with 100-grit sandpaper to increase the coefficient of friction (Fig. 1f). The gross weight of the instrumented apparatus was 172 g. Data was collected using a custom LabVIEW program (National Instrument, Austin, TX). Force signals were amplified and multiplexed using an ATI interface boxes (ATI Industrial Automation, Inc., Apex, NC), converged to 16-bit analog–digital converters (PCIe-6343, National Instrument, Austin, TX) and collected at a sampling frequency of 1000 Hz.

Experimental set-up, test protocol and representative signals of the reach-to-grasp performance. a A spherical retro-reflective marker placed a custom stand as the target for the reach-to-grasp kinematic task. b The mirror system for reach-to-grasp kinematic test. The reflective side of the mirror was facing the left alley and the grasping hand was behand the mirror in the right alley. c Protocol of reach-to-grasp kinematic test. d, e Trajectories in the horizontal plane (x–z plane) during reaching, grasping and returning under VF (d) and NVF (e) conditions from a representative AD patient. f The apparatus for reach-to-grasp kinetics. g The mirror system for reach-to-grasp kinetic test. h Protocol of reach-to-grasp kinetic test. i, j Grip force, load force and vertical position of the apparatus during precision grip under VF (i) and NVF(j) conditions from a representative AD patient

A mirror operating system was designed to block the visual feedback of the grasping hand and forearms from reach-to-grasp action [23]. The mirror was approximately 50 cm × 49 cm height and width, and 1 cm in thickness. After the mirror was in place, the space in front of the subject could be divided into two alleys. The reflective and the coating sides of the mirror are facing the left and right alleys, respectively. A marker-target for the kinematic test or a surrogate apparatus for the kinetic test was placed on the left alley, so that at the symmetric position with respect to the mirror a target or an apparatus for grasping was observable. Using this mirror system, the subject’s reaching right hand was behind the mirror in the right alley so that the visual supervision of the grasping hand and forearm could be blocked but the visual information about the target’s location was remained, which was designated as the without visual feedback (NVF) condition (Fig. 1b, g). By contrast, once the mirror system was removed, all the visual information about the target and the grasping hand and arm were available, which refers to the visual feedback (VF) condition. For one who completed the tests under the two visual conditions, the differences between the VF and NVF conditions should be mainly attributed to the effects of vision, rather than the other factors such as the differences in muscle strength between subjects. Comparison between the two visual conditions could allow to observe the AD-related sensorimotor deficits.

The subjects sat comfortably at a table, with the right elbow flexed approximately 90° in the parasagittal plane, the left hand naturally on the left side of body. The grasping target was rigidly fixed on the testing table, aligned with the subject’s right shoulder and at a distance of 35 cm in front of the subject. The right hand was placed on the start position of the table before each trial.

For each kinematic trial, the subject was required to reach and grasp the marker-target with the tips of the thumb and index finger following auditory cues for consecutive five times. After receiving an audible ‘go’ command, the subject reached with his or her right hand towards the virtual target. To minimize dwell near contact, the subject immediately returned the hand to the starting position on the third beep to complete the trial. The subject was instructed to pinch the target with the thumb and index finger as accurately and consistently as possible (Fig. 1c). For each kinetic trial, the subject was instructed to reach and grasp the apparatus with his or her thumb and index finger. After receiving an auditory cue, the subject lifted the apparatus vertically about 13 cm above the testing table, and maintained the apparatus in the air as stably as he or she could for 5 s. After receiving another auditory cue, the subject replaced the apparatus at the testing table and then returned his or her grasping hand to the initial position (Fig. 1h). The reach-to-grasp kinematic and kinetic tests were performed equally in both the VF and NVF conditions.

All the kinematic signals recorded by the motion capture system were filtered with a fifth-order Butterworth digital filter at a cutoff frequency of 5 Hz. The onset of the reaching was determined once the velocity of the moving hand exceeded 5 mm/s. We defined the grasping time as the duration from the onset of reaching to the timepoint when the hand returned to the initial position (Fig. 1d, e).

The spatial localizations of the contact points by the thumb and index finger for each subject were fitted by an ellipsoid, which included 95% of the pinch contact points by a principal component analysis (Fig. 2a, b). The volume (Vol) of the ellipsoid was computed as an estimation pinch accuracy i. A mean absolute error, defined as the Euclidean distance between the pinch contact location and the target, was calculated for each trial as follows:

$$MAE = \sqrt - x_ )^ + (y_ - y_ )^ + (z_ - z_ )^ }$$

(1)

The reach-to-grasp kinematic performances of representative AD and CN subjects with different visual conditions. Distributions of the contact points by the thumb and index finger from a representative AD (a) and an age- and gender-matched CN (b) in NVF. The ellipsoids are fitting spheres including 95% of contact points. The volumes of the ellipsoids indicate the precision of grasping with respect to the target. Trajectories in velocity-position phase diagram during reaching, grasping and returning from a representative AD and a CN subjects with VF (c) and NVF (d). The minimum-jerk and actual trajectories from a representative AD during reaching in VF (e) and NVF (f) conditions

where \(x_\), \(y_\) and \(z_\) are coordinates of the pinch contact positions and the \(x_\),\(y_\) and \(z_\) are the coordinates of the target. The pinch contact location was determined following the method developed in a previous study [28]. Specifically, using each nail marker-cluster as a reference for a 3-D coordinate system, a spherical model of the respective digit finger-pad was represented. A virtual ‘‘nail-point’’ was computed as a projection along the cluster stem to the dorsal surface of the nail and served as the center of the respective sphere. Using digital calipers, each digit thickness was measured as the transverse distance from dorsal surface to digit-pad prominence of the distal segment and served as the sphere radius. A pinch contact between the thumb and index finger onto the target was assumed to occur according to two criteria: (1) the surfaces of the representative spheres for the two digits were separated by a distance equal to or less than 10 mm (i.e., the diameter of the marker target), and (2) the inter-distance velocity between the sphere centers was less than 15 mm/s. The distance between digit sphere surfaces is denoted as “inter-pad” distance. In addition, a mean absolute error, defined as the Euclidean distance between the pinch contact location and the target.

A movement harmonicity was proposed to quantify the movement variability of reach-to-grasp kinematics. Previous studies have demonstrated that the movement trajectories during a self-paced reach-to-grasp performance normally presents as elliptic curves in a velocity-position phase diagram (the x-axis is distance between the reaching hand and the target and the y-axis is the velocity of the hand, Fig. 2c, d). The movement harmonicity can be computed as follows:

$$\begin \left\l} - R_ |}} }}} \\ = \frac }} }}} \\ = \frac }} }}} \\ \end } \right. \hfill \\ \hfill \\ \end$$

(2)

where the \(C_\) and \(A_\) are the circumference and area of an ideal ellipse whose major axis equals to the distance between the initial hand position and the target, and minor axis equals to the maximum velocity in the velocity-position phase diagram; the \(C_\) and \(A_\) are the circumference and area of the fitting ellipse of the movement trajectories in the velocity-position phase diagram.

A mathematic model [29] based on theory of dynamic optimization [30] was applied to quantify motor coordination during reach-to-grasp maneuver. Briefly, an objective function for motor coordination can be defined as follows:

$$\mathop \limits_ (\frac\int_^ }} x(t)}} }}} \right)^ + \left( y(t)}} }}} \right)^ } \right]} \, dt)$$

(3)

where \(x(t)\) and \(y(t)\) are the real-time coordinates of the hand in a planar motion, \(t_\) is the movement duration. A minimum-jerk trajectory algorithm was applied to estimate \(x(t)\) and \(y(t)\) that minimize the function [3]. The \(x(t)\) and \(y(t)\) can be expressed as fifth order polynomials as follows:

$$\left\c} + (x^ - x^ )( - 10(\frac }})^ + 15(\frac }})^ - 6(\frac }})^ )} \\ + (y^ - y^ )( - 10(\frac }})^ + 15(\frac }})^ - 6(\frac }})^ )} \\ \end } \right.$$

(4)

where the \((x^ ,y^ )\) and \((x^ ,y^ )\) are the initial and final coordinates of the reaching hand. The area between the trajectory of reaching hand and the curve formed by the (x(t), y(t)) in Eq. (4) of each trial serves as an indicator for motion coordination (Fig. 2e, f).

The apparatus was used to measure the forces (Fx, Fy and Fz) and torques (Tx, Ty and Tz) of the thumb and index finger, separately. All force and torque components were recorded simultaneously and then filtered using a fifth-order Butterworth low-pass filter with a cutoff frequency at 30 Hz (Fig. 1i, j). The grip force, GF, applied by the thumb and index finger, were the average of the two perpendicular forces. The load force, LF, was the summation of the vertical lifting forces applied by the thumb and index finger (Fig. 3a). Reach-to-grasp kinetics can be generally divided into five phases, including to reach, grasp, lift, hold and release the apparatus (Fig. 3b). The lifting phase can be further divided into a preload and a load subphases. The preload phase (Tpre) refers to the period from the moment when index finger and thumb first touched the object (the GF first exceeded 0.1 N for more than 2 s) to the onset of the load phase (the LF first exceeded 0.1 N) [31]. The load phase (Tload) refers to the onset of the load phase to the moment when the load force overcame the gravity so that the object started to move (Fig. 3b).

Force analysis for reach-to-grasp kinetic performance. a The force components applied by the thumb and index finger upon the apparatus; b the phases of reach-to-grasp kinetics and the GF and LF curves during grasping, lifting and holding the apparatus. The lifting phases can be further divided into the preload phase (in green) and the load phase (in pink)

The first derivative of GF versus time during the load phase was computed as grip force rate (GFR, Fig. 4a–d). A Gaussian function was used to fit the curve of GFR (Fig. 4e–h), and the root mean square errors (RMSEs) between the normalized GFR and the fitted Gaussian curve were calculated to quantify their differences. A continuous wavelet transform (CWT) with slow and fast bell-shaped functions (Mexican Hat waveform) was used to examine the time–frequency characteristics of the normalized GFR (Fig. 4i–l). The slow bell-shaped function indicates the components with lower frequency (or higher scale), reflecting the slowly changed GFR components. By contrast, the fast bell-shaped function indicates the components with higher frequency (or lower scale), which reflects the fast changes in GFR. To simplify the calculation, the slow bell-shaped component \(S(b)\) was defined as the average of the 5 scales of the slow bell-shaped function in formula (5). Similarly, the fast bell-shaped component \(F(b)\) was defined as the average of the 5 scales of the fast bell-shaped function. The percentage ratio \(R(b)\) was calculated as the division of the slow bell-shaped component to the sum of slow and fast bell-shaped components as specified in formula (5).

$$\left\l} \sum\limits_^ ,b)} } \\ \sum\limits_^ _ ,b)} } \\ } \times 100\% } \\ \end } \right.$$

(5)

Grip force rate analysis for reach-to-grasp kinetic performance. The grip force rates of representative subjects, including a CN subject in VF (a) and NVF (b) conditions, and an AD patient in VF (c) and NVF (d) conditions. Normalized grip force rates (GFR) and their fitted Gaussian functions for the CN subject in VF (e) and NVF (f) conditions and AD patient in VF (g) and NVF (h). The time–frequency spectrogram of the normalized GFR with continuous wavelet analysis for the CN subject in VF (i) and NVF (j) and the AD patient in VF (k) and NVF (l)

where \(a_\) = 15, 17.5, 20, 22.5, and 25 for the slow components and \(\tilde_\) = 70, 80, 90, 100, and 110 for the fast components. The average of \(R(b)\) during the load phase was calculated as a parameter for the statistical analysis.

The GF-LF coordination was estimated by computing a cross-correlation function based on the rates of change of the GF and LF. For each trial, the maximal coefficient of correlation (CC) and the time shifts (TS) were used to quantify the GF and LF coupling (Fig. 5a–d). The coefficient of variation (COV) which was defined as the ratio of the standard deviation of GF to the mean of GF during the first 5 s of the hold phase was used to quantify the variation of pinch force control (Fig. 5e). To determine the thumb and index finger tip positions on the manipulandum, the x and y coordinates of the center of pressure (COP) of each fingertip were measured during the hold phase. The COP data were fitted by an ellipse for the thumb and index finger (Fig. 5f), separately. The area of the ellipses in which 95% of the COP were located was computed as an estimate of the COP variability.

The GF-LF coordination and the center of pressure areas. The deviations of GF and LF of representative subjects with NVF for AD (a) and CN (b). The cross-correlation analysis based on the GF (c) and LF (d) rates of change. The maximal coefficient of correlation (CC) and the time shifts (TS) were used to quantify the GF and LF coupling. The coefficient of variation defined as the ratio of the standard deviation to the mean of GF of the hold phase (e). Distributions of the fingertip center of pressure and its area estimated by a fitted ellipse (f)

The validity of reach-to-grasp kinetic and kinematic parameters were examined with neuropsychological tests. Correlations analyses between the reach-to-grasp parameters and the scores of MMSE, MoCA, HAMA and the HAMD were performed for the AD group. The correlations were analyzed between each kinematic or kinetic parameters and each neuropsychological test scores, individually without consideration of multiple comparison. Only the correlations fulfilling statistically significance were retained as meaningful results.

All statistical analyses were performed using SPSS 25.0 (SPSS Inc., Chicago, IL). The kinematic and kinetic parameters were firstly examined for normality using a Kolmogorov–Smirnov test (K-S test). Analysis of variance (ANOVA) with repeated measures were employed to examine the differences of kinematic and kinetic parameters between the AD and CN groups as the between-subject factor across and the VF versus NVF conditions as the within-subject factor. Independent samples t-tests were applied to examine the difference in the kinematic and kinetic parameters between the AD and CN groups. Paired samples t-tests were applied to examine the effects of visual feedback for both the AD and CN groups. Correlation analyses between the neuropsychological test scores, including the MMSE, MoCA, HAMA, and HAMD, and the kinematic or kinetic parameters were further performed. A p-value of less than 0.05 was considered statistically significant.