Predicting Breath Hold Task Compliance From Head Motion

Abstract

Background: Cerebrovascular reactivity reflects changes in cerebral blood flow in response to an acute stimulus and is reflective of the brain's ability to match blood flow to demand. Functional MRI with a breath-hold task can be used to elicit this vasoactive response, but data validity hinges on subject compliance. Determining breath-hold compliance often requires external monitoring equipment.

Purpose: To develop a non-invasive and data-driven quality filter for breath-hold compliance using only measurements of head motion during imaging.

Study type: Prospective cohort.

Participants: Longitudinal data from healthy middle-aged subjects enrolled in the Coronary Artery Risk Development in Young Adults Brain MRI Study, N = 1141, 47.1% female.

Field strength/sequence: 3.0 Tesla gradient-echo MRI.

Assessment: Manual labelling of respiratory belt monitored data was used to determine breath hold compliance during MRI scan. A model to estimate the probability of non-compliance with the breath hold task was developed using measures of head motion. The model's ability to identify scans in which the participant was not performing the breath hold were summarized using performance metrics including sensitivity, specificity, recall, and F1 score. The model was applied to additional unmarked data to assess effects on population measures of CVR.

Statistical tests: Sensitivity analysis revealed exclusion of non-compliant scans using the developed model did not affect median cerebrovascular reactivity (Median [q1, q3] = 1.32 [0.96, 1.71]) compared to using manual review of respiratory belt data (1.33 [1.02, 1.74]) while reducing interquartile range.

Results: The final model based on a multi-layer perceptron machine learning classifier estimated non-compliance with an accuracy of 76.9% and an F1 score of 69.5%, indicating a moderate balance between precision and recall for the identification of scans in which the participant was not compliant.

Data conclusion: The developed model provides the probability of non-compliance with a breath-hold task, which could later be used as a quality filter or included in statistical analyses.

Level of evidence: 1: TECHNICAL EFFICACY: Stage 3.

Keywords: breath hold; cerebrovascular reactivity; data quality; fMRI.

FIGURE 1

Flow diagram for feature selection and model development.

FIGURE 2

Representative higher‐compliant and lower‐compliant scans including signals for head motion (FD) and respiratory belt (RSP). (A) FD representing head motion signal (solid black line, left y‐axis) and RSP (gray dashed line, right y‐axis) for a higher‐compliant participant with limited variation during BH (shaded area), $r_{\mathit{FD}-\mathit{RSP}}$ = 0.48. (B) FD and RSP for lower‐compliant participant with increased variation during BH, $r_{\mathit{FD}-\mathit{RSP}}$ = 0.32.

FIGURE 3

Example outliers defined by head motion metrics (FD) contrasting with manually‐rated task compliance based on respiratory belt (RSP). Shaded gray regions represent BH periods and manually‐rated RSP signals are depicted by dotted line. (A) FD and RSP traces from a scan that was manually‐rated as higher‐compliant but $R_{\mathit{\text{mean}}}$ = 2.6, which is lower than the outlier threshold of $R_{\mathit{\text{mean}}}$ < 3.2. (B) FD and RSP of a participant rated lower‐compliant but $R_{\mathit{\text{mean}}}$ = 6.39, which is greater than the outlier threshold of $R_{\mathit{\text{mean}}}$ > 4.2.

FIGURE 4

Performance diagnostics for the MLP and SVM models. (A) learning curves for MLP model depicting the model's performance over time for the training (blue) and validation loss (orange), showing effective MLP learning with minimal overfitting; (B) ROC curves derived from the first and second iterations of the MLP (blue, orange) and SVM (green, red) models, respectively, where the second iteration of the MLP model produced an AUC of 0.80; (C) confusion matrix derived from the MLP model with optimal threshold demonstrating high precision.

FIGURE 5

Model features as a function of compliance between training and application sets. Each panel shows individual model features included. Training set = manually rated compliance; Application set = model‐predicted compliance for scans that did not have RSP data. (A) Mean head motion during BH blocks expressed as framewise displacement (mm). (B) Standard deviation of head motion during BH blocks (mm), (C) ratio of mean head motion during free breathing compared to the mean head motion during BH blocks, (D) ratio of head motion standard deviation during free breathing compared to head motion standard deviation during BH blocks.