eyeris: Flexible, Extensible, & Reproducible Pupillometry Preprocessing

💻 eyeris DevOps Dashboard

Dive deeper into eyeris’ development and operational insights with our new eyeris DevOps Dashboard!

💡 Motivation

Despite decades of pupillometry research, many established packages and workflows unfortunately lack design principles based on (F)indability (A)ccessbility (I)nteroperability (R)eusability (FAIR) principles. eyeris, on the other hand follows a thoughtful design philosophy that results in an intuitive, modular, performant, and extensible pupillometry data preprocessing framework. Much of these design principles were heavily inspired by Nipype.

eyeris also provides a highly opinionated pipeline for tonic and phasic pupillometry preprocessing (inspired by fMRIPrep). These opinions are the product of many hours of discussions from core members and signal processing experts from the Stanford Memory Lab (Shawn Schwartz, Mingjian He, Haopei Yang, Alice Xue, and Anthony Wagner).

eyeris also introduces a BIDS-like structure for organizing derivative (preprocessed) pupillometry data, as well as an intuitive workflow for inspecting preprocessed pupillometry epochs within beautiful, interactive HTML report files (see demonstration below ⬇)! The package also includes gaze heatmaps that show the distribution of eye coordinates across the entire screen area, helping you assess data quality and participant attention patterns. These heatmaps are automatically generated in the BIDS reports and can also be created manually.

🚀 Feature Highlights

• 📦 Modular Design: Each preprocessing step is a standalone function that can be used independently or combined into custom pipelines.
• 🔍 Interactive Reports: Beautiful, interactive HTML reports that summarize preprocessing steps and visualize data quality.
• 🔄 Flexible Extensions: Easily create custom extensions to the preprocessing pipeline by writing your own functions and adding them to the pipeline.
• 📊 Data Quality Assessment: Automatically generated figures of each preprocessing step and its effect on the pupil signal (at the global and trial levels), as well as gaze heatmaps and binocular correlation plots to assess data quality and participant attention patterns.
• 🗂️ BIDS-like File Structure: Organizes preprocessed data using a BIDS-like directory structure that supports both monocular and binocular eye-tracking data.
• 📝 Logging Commands: Automatically capture all console output and errors to timestamped log files.

📖 Function Reference

Below is a table of all main eyeris functions, organized by feature, with links to their documentation and a brief description.

Feature	Function Documentation	Description
Pipeline Orchestration	glassbox()	Run the full recommended preprocessing pipeline with a single function call.
BIDSify	bidsify()	Create a BIDS-like directory structure for preprocessed data as well as interactive HTML reports for data and signal processing provenance.
Data Loading	load_asc()	Load EyeLink .asc files into an eyeris object.
Blink Artifact Removal	deblink()	Remove blink artifacts by extending and masking missing samples.
Transient (Speed-Based) Artifact Removal	detransient()	Remove transient spikes in the pupil signal using a moving MAD filter.
Linear Interpolation	interpolate()	Interpolate missing (NA) samples in the pupil signal.
Lowpass Filtering	lpfilt()	Apply a Butterworth lowpass filter to the pupil signal.
Downsampling	downsample()	Downsample the pupil signal to a lower sampling rate.
Binning	bin()	Bin pupil data into specified time bins using mean or median.
Detrending	detrend()	Remove slow drifts from the pupil signal by linear detrending.
Z-scoring	zscore()	Z-score the pupil signal within each block.
Confound Summary	summarize_confounds()	Summarize and visualize confounding variables for each preprocessing step.
Epoching & Baselining	epoch()	Extract time-locked epochs from the continuous pupil signal.
Plotting	plot()	Plot the pupil signal and preprocessing steps.
Gaze Heatmaps	plot_gaze_heatmap()	Generate heatmaps of gaze position across the screen.
Binocular Correlation	plot_binocular_correlation()	Compute correlation between left and right eye pupil signals.
Demo (Monocular) Dataset	eyelink_asc_demo_dataset()	Load a demo monocular recording EyeLink dataset for testing and examples.
Demo (Binocular) Dataset	eyelink_asc_binocular_demo_dataset()	Load a demo binocular recording EyeLink dataset for testing and examples.
Logging Commands	eyelogger()	Automatically capture all console output and errors to timestamped log files.
Database Storage	eyeris_db_collect()	High-performance database storage and querying alternative to CSV files.
Database Summary	eyeris_db_summary()	Get comprehensive overview of database contents and metadata.
Database Connection	eyeris_db_connect()	Connect to eyeris databases for custom queries and operations.
Custom Extensions	See vignette: Custom Extensions	Learn how to write your own pipeline steps and integrate them with eyeris.
Internal API Reference	See vignette: Internal API Reference	Comprehensive documentation of all internal functions for advanced users and developers.

For a full list of all functions, see the eyeris reference index.

📚 Tutorials

🌟 Start Here

• ✈ Getting Started: Complete (Opinionated) Pupillometry Pipeline Walkthrough
• 📁 Extracting Data Epochs and Exporting Pupil Data

👀 Pupil Data Quality Control

• 🔎 QC with Interactive Reports

💯 Advanced Topics

• 🫀 Anatomy of an eyeris Object
• 🛠 Building Your Own Custom Pipeline Extensions
• 🗄 Database Storage Guide: Scalable Alternative to CSV Files

📦 Package Installation

Stable release from CRAN

You can install the stable release of eyeris from CRAN with:

install.packages("eyeris")

or

# install.packages("pak")
pak::pak("eyeris")

Development version from GitHub

You can install the development version of eyeris from GitHub with:

# install.packages("devtools")
devtools::install_github("shawntz/eyeris", ref = "dev")

✏ Example

The glassbox() “prescription” function

This is a basic example of how to use eyeris out of the box with our very opinionated set of steps and parameters that one should start out with when preprocessing pupillometry data. Critically, this is a “glassbox” – as opposed to a “blackbox” – since each step and parameter implemented herein is fully open and accessible to you. We designed each pipeline step / function to be like a LEGO brick – they are intentionally and carefully designed in a way that allows you to flexibly construct and compare different pipelines.

We hope you enjoy! -Shawn

set.seed(32)

library(eyeris)
#> 
#> eyeris v2.1.1.9006 - Lumpy Space Princess ꒰•ᴗ•｡꒱۶
#> Welcome! Type ?`eyeris` to get started.

demo_data <- eyelink_asc_demo_dataset()

eyeris_preproc <- glassbox(
  demo_data,
  lpfilt = list(plot_freqz = FALSE)
)
#> ✔ [2025-08-09 16:20:18] [OKAY] Running eyeris::load_asc()
#> ℹ [2025-08-09 16:20:18] [INFO] Processing block: block_1
#> ✔ [2025-08-09 16:20:18] [OKAY] Running eyeris::deblink() for block_1
#> ✔ [2025-08-09 16:20:18] [OKAY] Running eyeris::detransient() for block_1
#> ✔ [2025-08-09 16:20:18] [OKAY] Running eyeris::interpolate() for block_1
#> ✔ [2025-08-09 16:20:18] [OKAY] Running eyeris::lpfilt() for block_1
#> ! [2025-08-09 16:20:18] [WARN] Skipping eyeris::downsample() for block_1
#> ! [2025-08-09 16:20:18] [WARN] Skipping eyeris::bin() for block_1
#> ! [2025-08-09 16:20:18] [WARN] Skipping eyeris::detrend() for block_1
#> ✔ [2025-08-09 16:20:18] [OKAY] Running eyeris::zscore() for block_1
#> ℹ [2025-08-09 16:20:18] [INFO] Block processing summary:
#> ℹ [2025-08-09 16:20:18] [INFO] block_1: OK (steps: 6, latest:
#> pupil_raw_deblink_detransient_interpolate_lpfilt_z)
#> ✔ [2025-08-09 16:20:18] [OKAY] Running eyeris::summarize_confounds()

Step-wise correction of pupillary signal

plot(eyeris_preproc, add_progressive_summary = TRUE)

Final pre-post correction of pupillary signal (raw ➡ preprocessed)

start_time <- min(eyeris_preproc$timeseries$block_1$time_secs)
end_time <- max(eyeris_preproc$timeseries$block_1$time_secs)

plot(eyeris_preproc,
  # steps = c(1, 5), # uncomment to specify a subset of preprocessing steps to plot; by default, all steps will plot in the order in which they were executed by eyeris
  preview_window = c(start_time, end_time),
  add_progressive_summary = TRUE
)
#> ℹ [2025-08-09 16:20:18] [INFO] Plotting block 1 with sampling rate 1000 Hz from
#> possible blocks: 1

#> ℹ [2025-08-09 16:20:18] [INFO] Creating progressive summary plot for block_1

#> ✔ [2025-08-09 16:20:19] [OKAY] Progressive summary plot created successfully!

plot_gaze_heatmap(
  eyeris = eyeris_preproc,
  block = 1
)

🗄 Database Storage: Scalable Alternative to CSV Files

eyeris includes powerful database functionality powered by DuckDB that provides a scalable, efficient alternative to CSV file storage. This is especially valuable for large studies, cloud computing, and collaborative research projects.

Why Use Databases?

🚀 Performance at Scale - Handle hundreds of subjects efficiently vs. managing thousands of CSV files - Faster queries: filter and aggregate at the database level instead of loading all data into R - Reduced memory usage: load only the data you need

💯 Cloud Computing Optimized - Reduce I/O costs on AWS, GCP, Azure - Single database file vs. thousands of CSV files for data transfer - Bandwidth efficient and cost-effective for large datasets

🔒 Data Integrity - Built-in schema validation prevents data corruption - Automatic metadata tracking and timestamps

Quick Start: eyeris Project Database Creation

Enable eyeris project database storage alongside or instead of CSV files:

bidsify(
  processed_data,
  bids_dir = "~/my_study",
  participant_id = "001",
  session_num = "01", 
  task_name = "memory_task",
  csv_enabled = TRUE,    # keep traditional BIDS-style CSV output files
  db_enabled = TRUE,     # but also create an eyeris project database
  db_path = "study_database"
)

bidsify(
  processed_data,
  bids_dir = "~/my_study",
  participant_id = "001", 
  session_num = "01",
  task_name = "memory_task", 
  csv_enabled = FALSE,   # skip CSV creation
  db_enabled = TRUE,     # cloud-optimized: Database only (no CSV files)
  db_path = "study_database"
)

Simple Data Extraction

Extract all your data with one function call:

# extract ALL data for ALL subjects
all_data <- eyeris_db_collect("~/my_study", "study_database")

# access specific data types
timeseries_data <- all_data$timeseries
confounds_data <- all_data$run_confounds

# targeted extraction: specific subjects and data types
subset_data <- eyeris_db_collect(
  "~/my_study", 
  "study_database",
  subjects = c("001", "002", "003"),
  data_types = c("timeseries", "epochs", "confounds_summary")
)

Database Overview and Management

# get a comprehensive database summary
summary <- eyeris_db_summary("~/my_study", "study_database")
summary$subjects      # all subjects in database
summary$data_types    # available data types  
summary$total_tables  # number of tables

# connect to eyeris database for custom operations
con <- eyeris_db_connect("~/my_study", "study_database")
# ... custom SQL queries ...
eyeris_db_disconnect(con)

💡 Pro Tip: Use csv_enabled = FALSE, db_enabled = TRUE for cloud computing to maximize efficiency and minimize costs.

📖 Complete Guide: See the Database Storage Guide for comprehensive tutorials, advanced usage, and real-world examples.

📁 BIDS-like file structure

eyeris organizes preprocessed data using a BIDS-like directory structure that supports both monocular and binocular eye-tracking data. The bidsify() function creates a standardized directory hierarchy with separate organization for different data types.

Monocular data structure

For single-eye recordings, data are organized in the main eye directory:

bids_dir/
└── derivatives/
    └── sub-001/
        └── ses-01/
            ├── sub-001.html
            └── eye/
                ├── sub-001_ses-01_task-test_run-01_desc-timeseries_eye.csv
                ├── sub-001_ses-01_task-test_run-01_desc-confounds.csv
                ├── sub-001_ses-01_task-test_run-01_epoch-stimulus_desc-preproc_pupil.csv
                ├── sub-001_ses-01_task-test_run-01_baseline-stimulus_desc-preproc_pupil.csv
                ├── sub-001_ses-01_task-test_run-01_events.csv
                ├── sub-001_ses-01_task-test_run-01_blinks.csv
                ├── sub-001_ses-01_task-test_run-01_summary.csv
                ├── sub-001_ses-01_task-test_run-01.html
                └── source/
                    ├── figures/
                    │   └── run-01/
                    │       ├── run-01_fig-1_deblink.jpg
                    │       ├── run-01_fig-2_detrend.jpg
                    │       ├── run-01_fig-3_interpolate.jpg
                    │       ├── run-01_fig-4_lpfilt.jpg
                    │       ├── run-01_fig-5_zscore.jpg
                    │       ├── run-01_gaze_heatmap.png
                    │       ├── run-01_detrend.png
                    │       └── run-01_desc-progressive_summary.png
                    └── logs/
                        └── run-01_metadata.json

Binocular data structure

For binocular recordings, data are organized into separate left and right eye subdirectories:

bids_dir/
└── derivatives/
    └── sub-001/
        └── ses-01/
            ├── sub-001-L.html
            ├── sub-001-R.html
            ├── eye-L/
            │   ├── sub-001_ses-01_task-test_run-01_desc-timeseries_eye-L.csv
            │   ├── sub-001_ses-01_task-test_run-01_desc-confounds_eye-L.csv
            │   ├── sub-001_ses-01_task-test_run-01_epoch-stimulus_desc-preproc_pupil_eye-L.csv
            │   ├── sub-001_ses-01_task-test_run-01_baseline-stimulus_desc-preproc_pupil_eye-L.csv
            │   ├── sub-001_ses-01_task-test_run-01_events_eye-L.csv
            │   ├── sub-001_ses-01_task-test_run-01_blinks_eye-L.csv
            │   ├── sub-001_ses-01_task-test_run-01_summary_eye-L.csv
            │   ├── sub-001_ses-01_task-test_run-01_eye-L.html
            │   └── source/
            │       ├── figures/
            │       │   └── run-01/
            │       └── logs/
            │           └── run-01_metadata.json
            └── eye-R/
                ├── sub-001_ses-01_task-test_run-01_desc-timeseries_eye-R.csv
                ├── sub-001_ses-01_task-test_run-01_desc-confounds_eye-R.csv
                ├── sub-001_ses-01_task-test_run-01_epoch-stimulus_desc-preproc_pupil_eye-R.csv
                ├── sub-001_ses-01_task-test_run-01_baseline-stimulus_desc-preproc_pupil_eye-R.csv
                ├── sub-001_ses-01_task-test_run-01_events_eye-R.csv
                ├── sub-001_ses-01_task-test_run-01_blinks_eye-R.csv
                ├── sub-001_ses-01_task-test_run-01_summary_eye-R.csv
                ├── sub-001_ses-01_task-test_run-01_eye-R.html
                └── source/
                    ├── figures/
                    │   └── run-01/
                    └── logs/
                        └── run-01_metadata.json

File naming convention

All files follow a consistent BIDS-like naming pattern:

• Timeseries data: desc-timeseries_eye (with _eye-L or _eye-R suffix for binocular data)
• Confounds: desc-confounds (with eye suffix for binocular data)
• Epochs: epoch-{event}_desc-preproc_pupil (with eye suffix for binocular data)
• Baselines: baseline-{event}_desc-preproc_pupil (with eye suffix for binocular data)
• Events: events (with eye suffix for binocular data)
• Blinks: blinks (with eye suffix for binocular data)
• Reports: HTML files with eye suffix for binocular data

Events and blinks data

The events and blinks CSV files contain the raw event markers and blink detection data as stored in the eyeris object:

Events file structure:

• block: Block/run number
• time: Timestamp of the event
• text: Raw event text from the ASC file
• text_unique: Unique event identifier

Blinks file structure:

• block: Block/run number
• stime: Start time of the blink
• etime: End time of the blink
• dur: Duration of the blink in milliseconds
• eye: Eye identifier (L/R for binocular data)

Key features

• Organized Structure: Clear separation between monocular and binocular data
• Consistent Naming: Standardized file naming across all data types
• Complete Documentation: HTML reports with preprocessing summaries and visualizations
• Quality Assessment: Gaze heatmaps and binocular correlation plots for data quality evaluation
• Reproducibility: Metadata files documenting preprocessing parameters and call stacks

📝 Logging eyeris commands with eyelogger()

The eyelogger() utility lets you run any eyeris command (or block of R code) while automatically capturing all console output and errors to timestamped log files. This is especially useful for reproducibility, debugging, or running batch jobs.

How it works:

• All standard output (stdout) and standard error (stderr) are saved to log files in a directory you specify (or a temporary directory by default).
• Each run produces two log files:
  • <timestamp>.out: all console output
  • <timestamp>.err: all warnings and errors

Usage

You can wrap any eyeris command or block of code in eyelogger({ ... }):

library(eyeris)

# log a simple code block with messages, warnings, and prints
eyelogger({
  message("eyeris `glassbox()` completed successfully.")
  warning("eyeris `glassbox()` completed with warnings.")
  print("some eyeris-related information.")
})

# log a real eyeris pipeline run, saving logs to a custom directory
log_dir <- file.path(tempdir(), "eyeris_logs")
eyelogger({
  glassbox(eyelink_asc_demo_dataset(), interactive_preview = FALSE)
}, log_dir = log_dir)

Parameters

• eyeris_cmd: The code to run (wrap in {} for multiple lines).
• log_dir: Directory to save logs (default: a temporary directory).
• timestamp_format: Format for log file names (default: "%Y%m%d_%H%M%S").

What you get

After running, you’ll find log files in your specified directory, e.g.:

20240614_153012.out   # console output
20240614_153012.err   # warnings and errors

This makes it easy to keep a record of your preprocessing runs and debug any issues that arise.

---

🙈 eyeris dependency graph

---

🤝 Contributing to eyeris

Thank you for considering contributing to the open-source eyeris R package; there are many ways one could contribute to eyeris.

We believe the best preprocessing practices emerge from collective expertise and rigorous discussion. Please see the contribution guidelines for more information on how to get started..

📜 Code of Conduct

Please note that the eyeris project is released with a Contributor Code of Conduct. By contributing to this project, you agree to abide by its terms.

💬 Suggestions, questions, issues?

Please use the issues tab (https://github.com/shawntz/eyeris/issues) to make note of any bugs, comments, suggestions, feedback, etc… all are welcomed and appreciated, thanks!

📚 Citing eyeris

If you use the eyeris package in your research, please consider citing our preprint!

Run the following in R to get the citation:

citation("eyeris")
#> To cite package 'eyeris' in publications use:
#> 
#>   Schwartz ST, Yang H, Xue AM, He M (2025). "eyeris: A flexible,
#>   extensible, and reproducible pupillometry preprocessing framework in
#>   R." _bioRxiv_, 1-37. doi:10.1101/2025.06.01.657312
#>   <https://doi.org/10.1101/2025.06.01.657312>.
#> 
#> A BibTeX entry for LaTeX users is
#> 
#>   @Article{,
#>     title = {eyeris: A flexible, extensible, and reproducible pupillometry preprocessing framework in R},
#>     author = {Shawn T Schwartz and Haopei Yang and Alice M Xue and Mingjian He},
#>     journal = {bioRxiv},
#>     year = {2025},
#>     pages = {1--37},
#>     doi = {10.1101/2025.06.01.657312},
#>   }