We use this program to analyze the response and gaze behavior of subjects when they play Go/NoGo task. We represent each subject using two-type of features: spatial and gaze-adjustment features.
Spatial features comprised of 24 index.
| No | ID | Detail |
|---|---|---|
| 1 | Go-positive | The percentage of Go response |
| 2 | Go-negative | The percentage of Go-negative response |
| 3 | NoGo-positive | The percentage of NoGo response |
| 4 | NoGo-negative | The percentage of NoGo-negative response |
| 5 | RT | The average of a subject response time |
| 6 | RT-var | The standard deviation of a subject response time |
| 7 | Trajectory-area | The gaze trajectory area |
| 8 | Velocity-avg | The average instantaneous velocity of subjects' gaze |
| 9 | Velocity-var | The standard deviation of the instantaneous velocity of subjects' gaze |
| 10 | Acceleration-avg | The average acceleration of subjects' gaze |
| 11 | Acceleration-var | The standard deviation of the velocity of subjects' gaze along the y-axis |
| 12 | Fixation-avg | The average of subjects' fixation time |
| 13 | Fixation-var | The standard deviation of subjects' fixation time |
| 14 | Distance-avg | The average of gaze distance |
| 15 | Distance-var | The standard deviation of gaze distance |
| 16 | Angle-avg | The average of gaze angle |
| 17 | Angle-var | The standard deviation of gaze angle |
| 18 | Distance-sen | Sample entropy of subjects' gaze distance |
| 19 | Angle-sen | Sample entropy of subjects' gaze angle |
| 20 | Velocity-sen | Sampe entropy of gaze velocity |
| 21 | Spatial-en | The entropy of subjects' gaze |
| 22 | Gaze-obj-en | The entropy of the distance between subjects' gaze and stimulus position |
| 23 | Gaze-obj-sen | Sample entropy of the distance between subjects' gaze and stimulus position |
| 24 | Gaze-obj-spe | Spectral entropy of the distance between subjects' gaze and stimulus position |
We analyzed the gaze-adjustment of participants for every type of response. Gaze-adjustment was the Euclidean distance between stimulus position and the subject's gaze when the stimulus appeared until it disappeared; its value ranged from 0 to sqrt(2). Since each gaze-adjustment's span differed, we represented each gaze-adjustment as Auto-regressive parameters. The model's lag L was set to five (average of AIC: 126.77), thereby each gaze-adjustment was represented by six variables.
We performed classification using AdaBoost with DecisionTree as the base classifier. The hyperparameter values of the AdaBoost algorithm and the classifier were optimized using grid-search. Three-fold cross validation was performed to evaluate the algorithm's performance.
- Analysis: codes to perform statistical analysis and classification
- Conf: constant variables (please modify the variables to match with your environment)
- DNNModels: Siamese neural network to discriminate ASD from typical ones (under progress)
- GazeAdjustment: codes to extract gaze adjustment features
- Spatial: codes to extract spatial features
- Testing: testing files
- Utils: library