TL;DR
This paper introduces and evaluates the NOS.E, a novel electronic nose (e-nose) designed for detecting volatile organic compounds (VOCs) associated with human decomposition and starvation, crucial for post-mass disaster victim identification. The e-nose demonstrates promising capabilities in differentiating these VOC profiles, offering a portable, cost-effective alternative to traditional search methods.
What Problem Does This Paper Address?
Mass disaster events are increasing in frequency, making the rapid and accurate location of victims paramount. While scent-detection dogs are currently the most effective method, they have limitations including short working times, potential for false positives, high costs, and an inability to convey detailed information. Existing analytical methods for VOC detection are often lab-based, expensive, and not portable. This research addresses the need for a field-deployable, cost-effective system that can detect and differentiate VOCs indicative of human remains and starvation, aiding in disaster victim identification.
Methods
The study utilizes a novel electronic nose, the NOS.E, equipped with an array of metal oxide semiconductor (MOS) sensors. These sensors were chosen for their known sensitivity to chemical classes commonly found in ante-mortem (starvation) and decomposition VOC profiles. The NOS.E system incorporates an automated airflow and control system to minimize environmental interference.
Key methodological aspects include:
- Analytical Standards: Fourteen analytical standards representing known ante-mortem and decomposition VOCs were used.
- Sample Preparation: An air dilution method was employed to create five different VOC concentrations for each analyte, ensuring a range of testing conditions.
- E-nose Testing Protocol: A three-phase sampling protocol (pre-conditioning, sampling, and recovery/cleaning) was implemented, with specific vacuum and sampling times optimized through experimental determination.
- Data Analysis: Sensor responses were analyzed using Python, employing automated peak detection and visualization tools. Principal Component Analysis (PCA) was used to visualize the collective sensor responses and assess the system's ability to differentiate analytes. For field trial data, Partial Least Squares-Discriminant Analysis (PLS-DA) was used to determine feature importance.
- Limit of Detection (LOD) Calculation: LOD was calculated using the residual standard deviation from calibration curves.
Datasets
The study did not use pre-existing datasets for training. Instead, it generated its own datasets through controlled experiments:
- Analytical Standards Dataset: Data was collected from analyzing fourteen different chemical compounds at various concentrations (0.2–119 ppm) using the NOS.E. This dataset was used to generate standard curves and assess the system's quantitative and qualitative performance.
- Simulated Mass Disaster Dataset: Data was collected from six donated human cadavers placed in a simulated disaster site. Samples were taken over 14 days post-placement from both areas with and without remains (control). This dataset was used to evaluate the system's in-field performance and its ability to differentiate known human remains from control samples.
Key Insights and Ideas
- Differentiating VOC Profiles: The NOS.E demonstrated a significant ability to differentiate between various chemical classes, including alcohols, hydrocarbons, ketones, and sulfur/nitrogen-containing compounds. This is crucial for distinguishing between ante-mortem (starvation) and decomposition VOC profiles.
- Sensor Importance: The OH, S, N, and Aromatics sensor (TGS 2602) emerged as a key contributor to the system's ability to differentiate analytes, both in laboratory tests and in the field trial.
- Low Detection Limits: The e-nose achieved an average LOD of 7.9 ppm, with some compounds detected at much lower concentrations, making it potentially useful for detecting victims even in challenging environments.
- Field Trial Success: In a simulated mass disaster, the NOS.E showed significant differences in sensor response between areas with human remains and control areas, particularly from day 3 post-placement onwards.
- Portability and Cost-Effectiveness: The NOS.E offers advantages in portability, shorter sampling times, and lower costs compared to traditional detection dogs and benchtop analytical instruments.
Key Results
- The NOS.E could detect and differentiate a range of VOCs prevalent in ante-mortem and decomposition profiles with an average LOD of 7.9 ppm.
- The system showed a significant difference between known human donor samples and control samples from day 3 post-mortem in the simulated disaster scenario.
- PCA analysis indicated good separation and clustering of analytes, particularly at the 1.6–11.9 ppm concentration range, suggesting the system's potential for distinguishing between different VOC profiles.
- The OH, S, N, and Aromatics sensor (TGS 2602) was identified as a significant feature for classification in the field trial.
- The system exhibited limitations in differentiating ethers and halogenated compounds due to the lack of specific sensors in the array.
Limitations and Future Work
- Quantification Challenges: While the NOS.E shows promise for qualitative detection, the R² values for most compounds were below 0.95, limiting its ability for accurate quantification of unknown concentrations or determining a Limit of Quantitation (LOQ).
- Sensor Drift and Fluctuations: Sensor drift and environmental factors (like wind) can affect performance in real-world scenarios, necessitating further research into algorithms to compensate for these issues.
- Sensitivity for Sub-Rubble Detection: While advantageous, the current sensitivity may still be insufficient for detecting victims buried deep under rubble, suggesting a need for improved sensor materials.
- Ethers and Halogens: The system struggled to differentiate ethers and halogenated compounds, indicating a need for sensor array modifications.
- Future Work: The authors suggest improving sensitivity through different sensor materials, conducting long-term repeatability studies, and performing additional field testing. Implementing machine learning algorithms (e.g., ANNs) is also highlighted as a critical next step to enhance selectivity, recognize complex VOC mixtures, and provide real-time alerts for untrained users in the field. Training the system on a wider range of ante-mortem and decomposition VOC profiles at different stages and with larger sample sizes is also recommended to account for human variability.
Summary updated