Identification of bodily fluid stains is crucial for forensic investigations. While current molecular detection methods provide high accuracy, their destructive sampling nature imposes significant limitations on trace samples by compromising sample integrity and subsequent short tandem repeat (STR) profiling. To date, no non-destructive method for fluid identification has been reported. This study introduces a novel portable electronic nose (e-nose) technology that facilitates non-invasive differentiation through the detection of volatile organic compounds (VOCs) in bodily fluids. Requiring only 3-4 minutes per test while preserving DNA integrity, this approach effectively distinguishes morphologically similar fluids such as blood and menstrual blood, offering an innovative solution for the non-destructive analysis of forensic body fluids. In this study, VOCs from 200 body fluid samples—including blood, saliva, semen, vaginal secretions (VS), and menstrual blood (MB)—were analyzed using electronic nose technology. Samples were collected via sterile swabs (n=100) and toilet paper (n=75). Radar plots indicated that sensor S7 (W1W) exhibited peak responses across both carriers. Linear discriminant analysis of the 175 samples revealed distinct clustering patterns (ANOSIM R = 0.088, p < 0.001). Subsequently, we employed four machine learning models—Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and Logistic Regression—for model development. RF demonstrated the best overall performance, while logistic regression performed best on samples collected with toilet paper. The prediction model based on RF showed an overall accuracy of 81.82% for the general model (n=175), with 100.00% accuracy for VS and 90.91% for blood. The toilet paper-based model achieved a higher overall accuracy of 92.00%, with 100.00% accuracy for saliva, MB, and VS. Feature contribution analysis revealed that W6S was the most important sensor in the general and swab-based models, whereas W3S was the most important in the toilet-paper-based model. Out-of-bag (OOB) error analysis indicated stable model generalization with ntree = 1000. The Kappa coefficient exceeded 0.7, the F1-score was above 0.8 for the toilet-paper-based model, and the macro-average AUC was greater than 0.9, demonstrating the robust consistency and discriminative ability of the models. The overall accuracy rate for the external validation set (n=25, sterile swab carriers) was 84.00% (100.00% for blood and saliva; 80.00% for semen and MB; 60.00% for VS). This study marks the first successful application of a portable electronic nose for the non-invasive identification of common bodily fluids, providing a new, low-cost, and easy-to-use non-destructive detection method for forensic analysis.

Authors

• Daijing Yu (School of Forensic Medicine, Shanxi Medical University, China)
• Niu Gao (School of Forensic Medicine, Shanxi Medical University, China)
• Tian Wang (School of Forensic Medicine, Shanxi Medical University, China)
• Liwei Zhang (School of Forensic Medicine, Shanxi Medical University, China)
• Jun Zhang (School of Forensic Medicine, Shanxi Medical University, China)
• Jiangwei Yan (School of Forensic Medicine, Shanxi Medical University, China)