Table 3.
Summary of identified studies using thermal imaging and VOC analysis as the screening modality.
| Study | Data source | ML classification methods | Performance metrics | Outcomes (best performing ML) |
|---|---|---|---|---|
| Thermal imaging | ||||
| Chakraborty et al.16 | FLIR T 650 SC long infrared (7.5–13 µm) camera | Support Vector Machine (SVM) | Accuracy | Accuracy 84.72 |
| Detection of volatile organic compounds (VOCs) | ||||
| Van de Goor et al.46 | ‘Aeonose’ electronic nose—using 3 micro-hotplate metal-oxide sensors to detect a range of VOCs in exhaled breath | Compression of 64 × 36 measurements per sensor, using Tensor Decompression (Tucker3-like). NN implemented through AeoNose software (Aethena software)—base architecture not specified |
Sensitivity Specificity Accuracy AUROC |
Sensitivity 84 Specificity 67 Accuracy 72 AUROC 0.850 |
| Mohamed et al.47 | ‘Aeonose’ electronic nose—using 3 micro-hotplate metal-oxide sensors to detect a range of VOCs in exhaled breath | Compression of 64 × 36 measurements per sensor, using Tensor Decompression (Tucker3-like). NN implemented through AeoNose software (Aethena software)—base architecture not specified |
Sensitivity Specificity Precision Accuracy AUROC |
Sensitivity 80 Specificity 77 Precision 67 Accuracy 79 AUROC 0.882 |
| Leunis et al.48 | ‘DiagNose’ electronic nose—12 metal-oxide sensors using four different sensor types: CH4, CO, NOx, Pt | Forward selection logistic regression |
Sensitivity Specificity AUROC |
Sensitivity 90 Specificity 80 AUROC 0.850 |
| Hakim et al.49 | ‘Nanoscale Artificial Nose’ (NA-NOSE) electronic nose—5 sensors based on gold nanospheres with tert-dodecanethiol, hexanathiol, 2-mercaptobenzoazole, 1-butanethiol, and 3-methyl-1-butanethiol ligands | Support vector machine (SVM) trains on principle components 1 and 2, following PCA of sensor measurements |
Sensitivity Specificity Accuracy |
Sensitivity 100 Specificity 92 Accuracy 96 |
| Mentel et al.18 | ‘BreathSpect’ device, utilising two fold separation using gas chromatography and mass spectrometry to detect VOCs | 2-Dimensional output from ‘BreathSpect’ device converted to integer arrays. Best classification performance using logistic regression | Accuracy | Accuracy 89 |