End-to-End Neural Network for Feature Extraction and Cancer Diagnosis of In Vivo Fluorescence Lifetime Images of Oral Lesions

Machine learning in point-of-care automated classification of oral potentially malignant and malignant disorders: a systematic review and meta-analysis

“…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 F 1 Sensitivity 87.5 Specificity 67.6 Precision 76.3 Accuracy 77.6 F 1 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.…”

“…S1 . Eight studies were found to have a high risk of bias across any of the 7 domains 2 , 16 , 21 , 22 , 26 , 28 , 30 , 35 . Within domain 1, 11% of studies were found to have high risk of bias, 26% low risk of bias, and 63% unclear risk of bias.…”

“…Four broad categories of methodologies were identified in POC detection of oral potentially malignant and malignant disorders: (1) classification based on clinical photographs (n = 11) 2 , 19 – 23 , 25 – 29 ; (2) in vivo imaging using intra-oral optical imaging techniques (n = 18) 15 , 17 , 30 , 31 , 33 – 35 , 37 – 45 , 50 ; (3) thermal imaging (n = 1) 16 ; (4) analysis of volatile organic compounds (VOCs) from breath samples (n = 5) 18 , 46 – 49 . Just 8 studies were published before 2015 15 , 34 , 37 , 38 , 44 , 48 – 50 .…”

“…Just 8 studies were published before 2015 15 , 34 , 37 , 38 , 44 , 48 – 50 . The majority of studies provided data on classification of OSCC vs healthy (n = 13) 16 , 18 , 19 , 23 , 31 , 33 , 38 , 42 , 43 , 46 – 49 , 8 studies provided data on OSCC/OPMD vs healthy 25 , 26 , 28 , 30 , 37 , 39 – 41 , 6 on OSCC/OPMD vs benign lesions 15 , 17 , 21 , 35 , 36 , 50 , 3 on OSCC vs benign 29 , 34 , 44 , 2 on OSCC vs other (healthy, benign and OPMD) 2 , 45 , 1 on OSCC/OPMD vs benign/healthy 20 , 1 on OPMD vs healthy 27 , and 1 on OPMD vs benign 22 .…”

“…This approach, using a VGG-M architecture pretrained on ImageNet, yielded an accuracy of 86.9%, with sensitivity of 85.0% and specificity of 88.7% 51 . Other approaches for achieving autofluorescence in vivo included a xenon lamp with monochromator and spectrograph 15 , multispectral digital microscopy 35 , time-domain multispectral endogenous fluorescence lifetime imaging FLIM 36 , N 2 laser 38 , confocal endomicroscopy (CFE) 33 , portable spectrophotometry 37 , 50 , and optical coherence tomography 45 . Notably, although in vivo and providing a prospect of POC detection, the approach taken by Aubreville et al of confocal laser endomicroscopy does require intra-venous administration of fluorescein prior to imaging and its utility as a POC detection tool may therefore be limited 33 .…”