Breath-based VOC analysis leveraging canine olfaction for multi-cancer detection: Insights from a 1000-sample study.

Abstract

1555 Background: Volatile organic compound (VOC) analysis is a validated approach for identifying disease-specific metabolic alterations through exhaled breath. The non-invasive and low-cost nature of breath sample collection makes it particularly suitable for large-scale cancer screening in resource-limited settings, such as those commonly found in the Global South. Canine olfaction has been demonstrated in prior controlled studies to detect VOCs with high accuracy across a range of pathologies, including malignancies. This study evaluates the performance of trained biomedical detection dogs in identifying multiple cancer types using VOC analysis and examines the integration of neurobehavioral data to support real-world diagnostic applications. Methods: A retrospective case-control study was conducted involving 1000 participants across three clinical sites in Hubli, India. Exhaled breath samples (n = 105 cancer-positive, n = 895 healthy controls) were collected using standardized protocols designed to maintain VOC integrity. Trained biomedical detection dogs analyzed these samples, with their behavioral responses recorded via motion sensors, video data, and electroencephalography (EEG) systems. A consensus-based decision framework was implemented to account for variability among individual dogs. Preliminary machine learning models were trained using the recorded neurobehavioral data to evaluate their potential for augmenting detection accuracy; however, these models remain in the validation phase. Results: The detection system demonstrated a sensitivity of 96% and a specificity of 100% across multiple cancer types in the test set, including oral, breast, esophageal, and cervical cancers. Sensitivity for early-stage cancers was 85%. The consensus-based approach among dogs enhanced reliability and minimized individual variability. Preliminary analysis of neurobehavioral data indicates potential for machine learning applications to refine diagnostic interpretation. Conclusions: Breath-based VOC analysis combined with canine olfaction demonstrates high accuracy in multi-cancer detection, including early-stage cancers. Its suitability for non-invasive and low-cost implementation, particularly in resource-constrained settings like the Global South, highlights its potential for addressing disparities in cancer screening access. Future research will focus on validating machine learning models and comparing the system's performance with existing diagnostic standards to further support global scalability and clinical adoption.