Abstract
Electronic nose (E-Nose) technology is gaining prominence as a tool for cancer diagnostics due to its ability to detect volatile organic compounds (VOCs) in bodily fluids. There has been no comprehensive study of the methodologic hurdles and overall diagnostic accuracy of E-Noses for cancer detection in blood samples. The aim of this investigation was to standardize a procedure for using a machine learning-based E-Nose to accurately diagnose five different cancers. This prospective (diagnostic/prognostic) study included 1001 newly diagnosed adult male and female participants with blood, brain, breast, liver, and lung cancers, as well as healthy controls. Blood samples were collected from all participants for complete blood counts, specific tumour markers testing, and E-Nose measurements. Sensor response patterns at the plateau region were used for training, testing, and machine learning cross-validation. Out of the final 550 participants, the E-Nose accurately categorized 100 to CLL, 50 to GBM, 150 to IDC, 50 to HCC, 50 to AC, and 150 to HC in agreement with specific tumour markers data; there were no false-positives or false-negatives. With an average area under the curve (AUC) of 1.0, the support vector machine had 100% accuracy, sensitivity, and specificity. Thus, the E-Nose had high diagnostic accuracy, sensitivity, and specificity in cancer detection in blood samples.