Evaluating the Use of the Infrascanner Device for Traumatic Brain Injury Triage in Uganda

Abstract

Background: Traumatic brain injury (TBI) is a major cause of morbidity and mortality in low- and middle-income countries, where limited access to computed tomography (CT) delays diagnosis of intracranial hematomas. The handheld InfraScannerTM detects supratentorial hematomas using near-infrared spectroscopy and performs well in high-resource settings. However, Ugandan studies have shown preserved sensitivity but markedly reduced specificity. We assessed whether adjusting the optical density difference (ODD) cutoff or targeting specific patient subgroups could improve diagnostic performance.Methods: We performed a secondary analysis of three prospective observational cohort studies: Study 1 at Duke University Hospital (Durham, NC, USA; December 2017–March 2018), Study 2 at Mbarara Regional Referral Hospital (MRRH), Southwestern Uganda (November 2020–May 2022), and Study 3 at MRRH and Mayanja Memorial Hospital (MMH), Southwestern Uganda (August 2024–February 2025). InfraScannerTM results were compared with CT across a range of ODD thresholds. Sensitivity, specificity, and AUC were calculated; optimal cutoffs were identified using Youden's Index. Logistic regression identified clinical predictors of false-positive results, enabling subgroup-level performance analysis. Results: At the factory cutoff (±0.20), Study 1 demonstrated high sensitivity (85.6%) and specificity (95.7%), with strong discrimination (AUC 0.96). In contrast, Study 2 showed sensitivity of 88.2% but specificity of only 33.1% (AUC 0.69), and Study 3 showed sensitivity of 89.8% with specificity of 29.3% (AUC 0.69). Threshold adjustment modestly improved specificity in the Ugandan cohorts (study 2 and 3) but did not restore discrimination to study 1 levels. Logistic regression identified scalp hematoma and moderate-to-severe TBI as independent predictors of false-positive results. Device performance was strongest among patients with mild TBI and no scalp hematoma (sensitivity 90%, specificity 50%; optimal cutoff: sensitivity 84%, specificity 68%). Conclusion: Reduced specificity of the InfraScannerTM in Uganda is unlikely to be explained solely by cutoff miscalibration. Although threshold adjustment modestly improved specificity, overall discrimination remained moderate. Diagnostic performance varied by clinical phenotype, with the strongest accuracy observed among patients with mild TBI and no scalp hematoma. These findings support phenotype-guided deployment of the device as a high-sensitivity triage tool in resource-limited settings rather than universal recalibration.