Human Assessment Module

This repository provides the code to support the creation of personalised machine learning models to quantify the cognitive performance and identify stress states, using physiological data. It is organised into four main components:

• Framework for Experimental Protocol and Data Collection
• Personalised Quantification of Cognitive Performance
• Personalised Classification of Stress
• Real-time Implementation

🧪 Framework for Experimental Protocol and Data Collection

The folder reaction_time_protocol contains the MATLAB experimental framework used to gather physiological and behavioral data, following the experimental design described in Rodrigues et al., 2018.

Protocol Overview

The experimental session includes the following tests:

• Baseline: Participants sit for 10 minutes to establish resting physiological signals.
• 2-Choice Reaction Time Task (CRTT): A selective-attention task, where participants identified either the large, global letter or the small, local letters of a hierarchically organized visual object. Their response time and correct/incorrect/missed answers were recorded.
• Trier Social Stress Test (TSST): A validated acute psychosocial stress paradigm involving public speaking and mental arithmetic tasks in front of an evaluative panel.

Order of Tasks:
Baseline → CRTT1 → TSST → CRTT2

Along the protocol, various psychological self-report scales were used and their results are saved in the created raw dataset.

Outputs

The experimental framework collects the following data per participant:

• ECG: Raw electrocardiogram (ECG) signal sampled during all protocol phases.
• VAS: Self-reported stress level using the Visual Analogue Scale.
• STAI_6items: Score from the 6-item short form of the State-Trait Anxiety Inventory, assessing current anxiety levels.
• Right_answers: The correct/expected responses for the CRTT task.
• Answers: Participant responses.
• Answer_reaction_time: Reaction time (in seconds) for each stimulus in the CRTT task.
• Answer_timing: Timestamps indicating when each visual stimulus was presented (used to align with physiological signals).

Dependencies

Psychtoolbox must be installed and tested before running the visual tasks.

🧠 Personalised Quantification of Cognitive Performance

Input

Each personalised model expects a JSON file as input containing rows of extracted physiological features. The expected structure is as follows:

File Type: .json

Keys:

Protocol Variables

Key Name	Description
Test_phase	The phase of the experiment (e.g., baseline, CRTT1, TSST, CRTT2).
ECG	Raw ECG signal, 1 lead & Fs = 500 HZ.
VAS	Visual Analogue Scale rating.
STAI_6items	State-Trait Anxiety Inventory (6-item version) score.
Accuracy	Percentage of correct responses in a cognitive task.
Reaction_time	Response time in seconds.
RT_std	Standard deviation of reaction times.
Cognitive_performance	Composite score of cognitive task performance.

_{Cognitive Performance = Accuracy / (Average Reaction Time × Reaction Time Std Dev)}

ECG Waveform Features

Key Name	Description
p_wave_duration	Duration of the P wave (ms).
pr_interval	Time from the start of the P wave to the start of the QRS complex (ms).
pr_segment	Time between the end of the P wave and the start of the QRS complex (ms).
qrs_duration	Duration of the QRS complex (ms).
qt_interval	Time between the start of the Q wave and the end of the T wave (ms).
st_segment	Segment between the QRS complex and the T wave (ms).
st_interval	Time between the J point and the end of the T wave (ms).
t_wave_duration	Duration of the T wave (ms).
tp_segment	Time from the end of the T wave to the start of the next P wave (ms).
rr_interval	Time between two consecutive R-peaks (ms).

Heart Rate Variability (HRV) Metrics

Key Name	Description
mean_nni	Mean of normal-to-normal (NN) intervals (ms).
sdnn	Standard deviation of NN intervals (ms).
sdsd	Standard deviation of successive differences between NN intervals (ms).
nni_50	Number of pairs of successive NN intervals differing by more than 50 ms.
pnni_50	Percentage of NN50 count divided by the total number of NN intervals.
nni_20	Number of pairs of successive NN intervals differing by more than 20 ms.
pnni_20	Percentage of NN20 count divided by the total number of NN intervals.
rmssd	Root mean square of successive differences (ms).
median_nni	Median of NN intervals (ms).
range_nni	Range of NN intervals (max-min) (ms).
cvsd	Coefficient of variation of successive differences.
cvnni	Coefficient of variation of NN intervals.

Heart Rate Metrics

Key Name	Description
mean_hr	Mean heart rate (beats per minute).
max_hr	Maximum heart rate recorded.
min_hr	Minimum heart rate recorded.
std_hr	Standard deviation of heart rate.

Frequency-Domain HRV Metrics

Key Name	Description
lf	Low-frequency power (ms²).
hf	High-frequency power (ms²).
lf_hf_ratio	Ratio of LF to HF power.
lfnu	Low-frequency power in normalized units.
hfnu	High-frequency power in normalized units.
total_power	Total spectral power of HRV (ms²).
vlf	Very low-frequency power (ms²).

Output Details

• Running the command will generate a model based on the provided dataset.
• Users must enter a unique code to the model when prompted.
• The model will be saved as a .pkl file in the cognition_personalised_models folder within the working directory. The file will follow this naming convention: cognition_personalised_models/model_<user_code>.pkl

Running the Code

To execute the algorithm, ensure you are in the root directory of the repository. The script should be run using the following command:

python -m code.personalisation_algorithms.cognition_model_personalisation

Upon execution, you will be prompted to select a file containing the dataset.

⚠ Attention: The input dataset must follow the required structure, though no example is currently provided, as the data used for testing is private.

⚡Personalised Classification of Stress

Input

Each personalised model expects a JSON file as input containing rows of extracted physiological features. The expected structure is the same presented for the cognition performance algorithm. See the "Personalised Quantification of Cognitive Performance" input description for detailed field structure.

Output Details

• Running the command will generate a set of models from the provided dataset and a CSV with the performance weights.
• Users must enter a unique code to the model when prompted.
• The outputs will be saved in the stress_personalised_models/Controller_<user_code> folder within the working directory. The model files will follow this naming convention: stress_<model_name>_C<user_code>.pkl, while the CSV follows:model_weights_C<user_code>.csv

Running the Code

To execute the algorithm, ensure you are in the root directory of the repository. The script should be run using the following command:

python -m code.personalisation_algorithms.stress_model_personalisation

Upon execution, you will be prompted to select a file containing the dataset.

⚠ Attention: The input dataset must follow the required structure, though no example is currently provided, as the data used for testing is private.

⚙️ Real-time Implementation

The real-time implementation loads personalised cognition and stress models based on a controller_id and applies them to updated physiological feature data, which is read every 60 seconds in a continuous loop.

Input

On the current code version, an example CSV file: example_ecg_features.csv with physiological features is uploaded. The input must include all the ECG features used for "Personalised Quantification of Cognitive Performance":

• ECG Waveform Features
• Heart Rate Variability (HRV) Metrics
• Heart Rate Metrics
• Frequency-Domain HRV Metrics

But should exclude the Protocol Variables, which are only available during training.

⚠️ This should be replaced with a real-time data stream , using the same feature structure as the example CSV.

The appropriate models are automatically selected based on the input controller_id.

Output details

The script prints:

• Cognition model prediction and explainability results

• Stress model prediction and explainability results

Results are printed to the console in each 60-second cycle.

In the future, the output can be extended to automatically send results to an API endpoint, store them in a database, or feed a real-time dashboard interface.

Running the Code

To execute the algorithm, ensure you are in the root directory of the repository. The script should be run using the following command:

python -m code.main

Dependencies

MATLAB Dependencies (for reaction_time_protocol)

• MATLAB R2018a or later
• Psychtoolbox-3
• Signal Processing Toolbox

Python Dependencies (for model training and real-time inference):

• Python 3.x
• Required Python libraries listed in requirements.txt

Install required Python libraries using:

pip install -r requirements.txt

License

This project is licensed under the MIT License.