Table 2.
Summary of identified studies using optical imaging as the screening modality.
| Study | Data source | ML classification methods | Performance metrics | Outcomes (best performing ML) |
|---|---|---|---|---|
| Optical imaging | ||||
| Uthoff et al.30 | Custom smartphone-based dual modality device capable of both white light and autofluorescence imaging | NN based on VGG-M architecture, pre-trained on ImageNet |
Sensitivity Specificity Precision NPV Accuracy AUROC |
Sensitivity 85.0 Specificity 89.0 Precision 0.88 NPV 0.85 Accuracy 86.9 AUROC 0.91 |
| Song et al.17 | Smartphone-based intraoral imaging system with custom WL probe | NN based on VGG19 architecture, pre-trained on ImageNet | Accuracy | Accuracy 85.6 |
| Chan et al.31 | VELscope device32 | Classification based ResNet or Inception architecture, using either a fully convolutional network or feature pyramid network |
Sensitivity Specificity |
Sensitivity 98.0 Specificity 88.0 |
| Aubreville et al.33 | Confocal Laser Endomicroscopy images of oral cavity following IV fluorescein. Images extracted from IO videos. CystoFlex UHD and Coloflex UHD as imaging devices | Used untrained LeNet-5 architecture with patch probability fusion, whole image classification using pre-trained Inception V3 CNN and random forest classifier. Best performance using LeNet-5 |
Sensitivity Specificity Accuracy AUROC |
Sensitivity 86.6 Specificity 90.0 Accuracy 88.3 AUROC 80.7 |
| De Veld et al.15 | Xe lamp with monochromator for illumination, a spectrograph and custom set of long-pass and short-pass filters | NN with base architecture not specified; single hidden layer between input and output | AUROC | AUROC 0.68 |
| Roblyer et al.34 | Multispectral digital microscope (MDM), measuring white light reflectance, autofluorescence, narrow band reflectance and cross-polarised light | Linear discriminant analysis |
Sensitivity Specificity AUROC |
Sensitivity 93.9 Specificity 98.1 AUROC 0.981 |
| Caughlin et al.35 | Multispectral autofluorescence lifetime imaging (maFLIM) endoscopy | Bespoke neural network using a shared encoder and separate paths for signal reconstruction and classification; classification on pixel-pixel basis |
Sensitivity Specificity Precision Accuracy F1 |
Sensitivity 87.5 Specificity 67.6 Precision 76.3 Accuracy 77.6 F1 0.80 |
| Jo et al.36 | Time-domain multispectral FLIM rigid endoscope. Emission spectral collected for collagen, NADH, FAD | Quadratic discriminant analysis |
Sensitivity Specificity AUROC |
Sensitivity 95 Specificity 87 AUROC 0.91 |
| Francisco et al.37 | Portable spectrophotometer with two solid state lasers; a diode emitting at 406 nm and a double frequency neodymium 523 nm as excitation source | Compared naïve bayes, k-Nearest Neighbours and decision tree. Decision tree provided best performance |
Sensitivity Specificity Accuracy |
Sensitivity 87.0 Specificity 91.2 Accuracy 87.0 |
| Wang et al.19 | Fibre optics-based flurospectrometer, using Xe lamp with monochromator as excitation source | Partial least squares combined with artificial neural network—neural network with single hidden layer |
Sensitivity Specificity Precision |
Sensitivity 81.0 Specificity 96.0 Precision 88 |
| Majumder et al.38 | N2 laser as excitation source | Relevance Vector Machine (RVM) |
Sensitivity Specificity AUROC |
Sensitivity 91 Specificity 95 AUROC 0.9 |
| Huang et al.39 | VELscope device | Quadratic discriminant analysis |
Sensitivity Specificity |
Sensitivity 92.3 Specificity 97.9 |
| Duran-Sierra et al.40 | Multispectral autofluorescence lifetime imaging endoscopy (maFLIM); preferential excitation of NADH and FAD | Best performance using ensemble approach of support vector machine and quadratic discriminant analysis |
Sensitivity Specificity F1 AUROC |
Sensitivity 94.0 Specificity 74.0 F1 0.85 AUROC 0.81 |
| Jeng et al.41 | VELscope device | Used both linear discriminant analysis and quadratic discriminant analysis |
Sensitivity Precision Accuracy F1 AUROC |
Sensitivity 92.0 Precision 0.86 Accuracy 86.0 F1 0.88 AUROC 0.96 |
| Huang et al.42 | Custom autofluorescence device, comprising two LED continuous wave lamps, for preferential imaging of NADH and FAD | Quadratic discriminant analysis |
Sensitivity Specificity |
Sensitivity 94.6 Specificity 85.7 |
| Kumar et al.43 | Custom portable autofluorescence device using collimating lens and bream splitter; 405 nm dioxide for excitation | Dimensionality reduction using PCA, before Mahalanobis distance classification on first 11 PCs |
Sensitivity Specificity Accuracy |
Sensitivity 98.7 Specificity 100 Accuracy 98.9 |
| Rahman et al.44 | Custom portable imaging system composed of modified headlamp system capable of both autofluorescence imaging and reflectance imaging | Linear discriminant analysis |
Sensitivity Specificity AUROC |
Sensitivity 92.0 Specificity 84.0 AUROC 0.913 |
| James et al.45 | Use of a spectral-domain Optical Coherence Tomography (OCT) system consisting of a 2D scanning long GRID rod probe with a centre wavelength of 930 nm | Use of 14 artificial neural networks for feature extraction, followed by support vector machine for classification. Best performance using DenseNet-201 and NASNetMobile in delineating OSCC from others |
Sensitivity Specificity PPV NPV Accuracy |
Sensitivity 86.0 Specificity 81.0 PPV 51.0 NPV 96.0 Accuracy 81.9 |