Detection of Δ9-Tetrahydrocannabinol Impairment Using Resting-State Functional Near-Infrared Spectroscopy: A Randomized Clinical Trial.

Summary

A brain-imaging technique called resting-state functional near-infrared spectroscopy (fNIRS) detected THC-induced impairment more accurately and with far fewer false positives than standard behavioral field sobriety tests (FSTs). In a double-blind, randomized crossover trial, 183 regular cannabis users (average age 25, half female) received either a single oral dose of synthetic THC or a placebo. fNIRS scans of the prefrontal cortex, taken at rest and during a working-memory task, were used to train machine-learning models to identify clinically determined intoxication. The fNIRS classifier achieved 90% accuracy and a 5% false-positive rate, compared with 69% accuracy and a 34% false-positive rate for FSTs. The findings suggest that resting-state fNIRS may provide a more reliable, objective method for detecting cannabis-related driving impairment.

Study at a glance

Abstract

The primary psychoactive compound in cannabis, ∆9-tetrahydrocannabinol (THC) induces intoxication and functional impairment, raising safety concerns in driving. The traditional impairment detection method, behavioral field sobriety tests (FSTs), are subject to bias. To determine whether resting-state functional near-infrared spectroscopy (fNIRS) can detect THC-related impairment with greater accuracy and a lower rate of false positives than FSTs. This double-blind, randomized, crossover trial was conducted from January 2017 to January 2021 at a single site. Eligible participants were adults aged 18 to 55 years who used cannabis. Analyses were performed from November 2024 to November 2025. Participants received a single oral dose of synthetic THC (range, 5-80 mg) intended to induce intoxication or placebo in separate visits. fNIRS scans were acquired before and approximately 100 and 200 minutes after study drug administration to assess prefrontal cortex responses at rest and during a working memory task. Machine learning models trained on fNIRS data were then used to identify clinically determined THC-induced impairment. The primary outcome of this study was accuracy of THC-induced impairment classification using fNIRS data as compared with an FST. Model performance was quantified using false-positive rate, precision, recall, F1 score, and area under the receiver operating curve (ROC-AUC). A total of 183 participants (mean [SD] age, 25.3 [6.3] years; 90 [49.2%] female) who used cannabis for a median (IQR) of 6.5 (4-7) days per week completed at least 1 study visit. fNIRS data collected during rest produced a classifier for impairment, with an ROC-AUC of 0.87 (95% CI, 0.83 to 0.91), accuracy of 0.90 (95% CI, 0.88 to 0.92), and false-positive rate of 0.05 (95% CI, 0.04 to 0.07), using clinical impairment assessment as ground truth. The FST showed an ROC-AUC of 0.75 (95% CI, 0.74 to 0.76), accuracy of 0.69 (95% CI, 0.67 to 0.71), and a false-positive rate of 0.34 (95% CI, 0.32 to 0.36). fNIRS performed significantly better than the FST in precision (difference = 0.23; 95% CI, 0.14 to 0.33; P < .001), accuracy (difference = 0.15; 95% CI, 0.10 to 0.19; P < .001), false-positive rate (difference = -0.25, 95% CI, -0.31 to -0.20; P < .001), and ROC-AUC (difference = 0.08; 95% CI, 0.01 to 0.14; P = .005). In this crossover trial of THC vs placebo, THC intoxication produced prefrontal cortex activation patterns detectable with resting state fNIRS neuroimaging, producing a neural signature of THC-induced impairment that was superior to FSTs for individual-level impairment identification. These findings lay the groundwork for further exploration of fNIRS as a tool for detecting impairment. ClinicalTrials.gov Identifier: NCT03655717.