New in AFNI's physio_calc.py (for FMRI physio regressors): QC images, reports and interactive mode

1690 

Abstract Submission 

Peter Lauren¹, Daniel Glen², Richard Reynolds¹, Joshua Dean¹, Daniel Handwerker³, Paul Taylor¹

¹National Institute of Mental Health, Bethesda, MD, ²NIMH, Bethesda, MD, ³Section on Functional Imaging Methods, NIMH, Bethesda, MD

Peter Lauren, PhD  
National Institute of Mental Health
Bethesda, MD

Daniel Glen  
NIMH
Bethesda, MD

Richard Reynolds  
National Institute of Mental Health
Bethesda, MD

Joshua Dean  
National Institute of Mental Health
Bethesda, MD

Daniel Handwerker, PhD  
Section on Functional Imaging Methods, NIMH
Bethesda, MD

Paul Taylor  
National Institute of Mental Health
Bethesda, MD

FMRI's BOLD signal includes both neuronal and non-neuronal contributions. Breathing and heart rate both have strong influences on blood oxygenation levels, and these sources are unlikely to be turned off during any in vivo scan. Therefore, it can be helpful to include measures of these biological phenomena within an FMRI model to more fully account for underlying signal contributions.

The AFNI software package [1] has recently developed the physio_calc.py program [2] to take physiological time series (such as breathing and heart rate measures) acquired during a scan session and create regressors for single subject analysis, such as within afni_proc.py or other pipeline tools. The program estimates both slicewise RETROICOR regressors [3], as well as respiration volume per time (RVT) [4]. These estimates depend on several time series analysis steps, such as peak and trough estimation across the curve, as well as local phase and amplitude. We describe new features in physio_calc.py that allow users to: reduce the effects of outliers and bad values in the time series; efficiently check and interactively correct initial estimates; see useful quality control (QC) images; and have quantitative reviews to help validate and check the underlying calculations.

The physio_calc.py program is written in Python and distributed within AFNI's afnipy module. It contains several automatic checks for corrupted data within input time series, such as NaN and missing values. Outlier checks, as well as replacement of potential "bad" values such as zeros that might be recorded erroneously, can also be enabled. A multistep algorithm is used to estimate peak and trough locations, including bandpassing, filtering and local refinement. A set of slicewise regressors for an accompanying FMRI time series is calculated, as well as RVT measures derived from the phases of the respiratory peaks and troughs.

Fig 1A shows an example of the QC output for respiratory time series processing, where estimated peaks and troughs (triangles) are shown with the input physio time series. The time interval over which the FMRI dataset was collected has a white background, and the remaining regions are helpful as boundary conditions. The top of each subplot contains color-coded rectangles for each interpeak interval: near-median values are white, while shorter ones are bluer and longer ones are redder. These can highlight either physio-based patterns within the time series, or potential QC issues (e.g., an erroneous or missing peak). The same appears at the bottom of each panel when troughs have been estimated. The user can add a "do_interact" option to have an interactive window open, to easily add, remove or shift points. All changes are saved immediately.

Fig 1B shows part of the review output text for each physio time series. This can easily be combined using AFNI into a spreadsheet, and queried to find outliers. This quantitative QC is helpful in summarizing group properties and/or finding data issues.

Fig 2 shows QC images of the computed physio regressors, for slicewise RETROICOR estimates and volume-wise RVT regressors. These regressors can be added directly into an FMRI processing pipeline stream, e.g., using afni_proc.py.

Physio regressors can play an important part in FMRI processing, but they rely heavily on how they are processed. To date, little formal attention has been paid to the QC of these physio inputs and their computed regressors. They are highly susceptible to noise and distortion (e.g., due to pulse oximeter finger movement), and must be checked for accuracy. AFNI's physio_calc.py contains several features to provide efficient QC images and quantitative reporting. It also has a simple interface for fixing any features within the estimates. Careful QC of all processing steps (both visual and quantitative) is key to all FMRI processing and sub-step processing [5], so these features should be considered vital in all analyses.

Classification and Predictive Modeling

Exploratory Modeling and Artifact Removal ¹

Methods Development ²

Computing

Data analysis

Design and Analysis

FUNCTIONAL MRI

Modeling

NORMAL HUMAN

Open Data

Open-Source Software

Workflows