Dataset for paper: Decoding multi-joint hand movements from brain signals by learning a synergy-based neural manifold

## Dataset and File Structure Description The dataset is organized into three folders: **NeuralData**, **KinData**, and **Info**. - **NeuralData**: Contains invasive brain signal recordings for each session. - **KinData**: Includes joint angle trajectories for 11 hand movements. - **Info**: Holds session metadata. The dataset consists of 3 sessions, each collected on a separate experimental day. Each session contains the following: 1. A **NeuralData** file in MATLAB `.mat` format. 2. An **Info** file in YAML format. The filenames indicate both the collection date and session number. For example, ‘**2022-06-27_Session_1.mat**’ refers to the first session, collected on June 27, 2022. Each session includes 11 different hand movements, with joint angle trajectories stored in the **KinData** folder. Filenames specify the gesture class and movement duration. For instance, ‘**gesture_data_901_2400.mat**’ represents the 901 gesture class, with an execution time of 2400 ms (2.4 seconds). ### Key Metadata Description ### **NeuralData Files**: - **N**: Number of trials - **C**: Number of neural channels - **T**: Number of time bins - **data**: A 3D matrix (N-T-C), representing neural activity across all trials - **target_label**: A vector of size N, indicating the hand target index for each trial (coded from 901 to 911) - **target_no**: A vector of size N, indicating the hand target index for each trial (coded from 0 to 10) - **len_trial**: A vector of size N, specifying the number of time bins for each trial ### **Info Files**: - **start_bin**: Bin index for the movement start - **end_bin**: Bin index for the movement end - **kin_data**: Length of the movement duration (e.g., 2400 denotes 2.4 seconds) - **num_block**: Number of blocks in the session - **num_movement**: Total number of movements ### **KinData Files**: - **D**: Number of motion dimensions (hand joints) - **M**: Number of target gestures (hand movements) - **T**: Number of time bins - **angle_base**: A vector of size M, representing the initial joint angle for each gesture, used as the base angle for all movements - **rotation_angle**: A matrix of size T-D, showing the joint rotation angle trajectory during movement - **rotation_axis**: A matrix of size T-Dx3, indicating the rotation axis for each joint in three-dimensional space **Note**: We recorded 16 joint angles during movement, consisting of one wrist joint and 15 hand joints. The analysis focuses on the kinematics of the 15 hand joints, using only the last 15 dimensions of the rotation angle data.