Blue-White Veil Classification of Dermoscopy Images Using Convolutional Neural Networks and Invariant Dataset Augmentation.

In the dermoscopy, the Three-Point Checklist of Dermoscopy and the Seven-Point Checklist are proved to be sufficient screening methods in the skin lesions assessments checking by dermatology expert. In the methods, there is a criterion of blue-whitish veil appearance within the lesion and it can be classified using CNN classifiers. In the paper, we show the results of CNN application to the problem of the assessment of whether the blue-white veil is present or absent within the lesion. We build the neural network with the help of the available VGG19, Xception and Inception-ResNet-v2 pretrained convolutional neural networks, trained, validated and tested on the prepared images taken from the PH2, using the invariant dataset augmentation. The original authors’ approach using the defined invariant dataset augmentation for expanding the test set by seven copies invariantly transformed from original images shows that the classification characteristics like accuracy and true positive rate as well as the F1 and MCC tests can be much higher (5–20%) than using only original images. In the paper, the confusion matrix parameters result in: 98–100% accuracy, 98–100% true positive rate, 0.0–2.3% false positive rate, tests F1 = 0.95 and MCC = 0.95 as well as AUC value close to 1. That general approach can provide higher results while using CNN networks in other disciplines not only in dermatology and dermoscopy.