Validation of an electronic-nose sensor array in detecting C. difficile and differentiating toxin-positive from toxin-negative stool samples
Year: 2026
Type: preprint
Abstract: Objective To evaluate the specimen-level performance of a multichannel electronic nose (e-nose) sensor array in classifying stool specimens into relevant Clostridioides difficile categories. Design Retrospective single-center validation study with reference to polymerase chain reaction (PCR) and C. difficile toxin testing. Participants A convenience sample of 31 discarded stool specimens with PCR and confirmatory toxin testing. Categorized into PCR-positive (+)/toxin+, PCR+/toxin-negative (–), and PCR–. Methods Specimens were sampled in controlled headspace by two e-nose sensor arrays in 10-minute sessions with 5-second sampling intervals. The primary endpoint was one-versus rest classification performance (sensitivity, specificity, positive predictive value, and accuracy) into three classes: PCR+/toxin+, PCR+/toxin–, and PCR–. Data were normalized and the region of interest was extracted for training and testing. Random forests were utilized for classification of normalized data with a test-train split at the specimen-level. Results There were 457 unique 10-minute sampling events (147 PCR+/toxin+, 183 PCR+/toxin–, 127 PCR–). Sensors 1 and 2 captured 20,331 and 23,538 unique 5-second exposures within the region of interest, respectively. Testing data classification accuracy was 0.79 [0.78, 0.81] and 0.84 [0.83, 0.85] for Sensors 1 and 2. Sensitivity for PCR+/toxin+ exposures was 0.82 [0.79, 0.84] and 0.94 [0.92, 0.95], and specificity was 0.91[0.90, 0.93] and 0.93 [0.92, 0.94] for Sensors 1 and 2. Conclusion An e-nose sensor array partially differentiated PCR-classified C. difficile specimen classes at the exposure level under headspace-controlled conditions. Performance was strongest for PCR+/toxin+ exposures. Findings support further prospective validation in the hospital room setting.
Source: medRxiv
Language: English
Subfield: Biomedical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7220892453