Automatic Diagnosis of Early-Stage Oral Cancer and Precancerous Lesions from ALA-PDD Images Using GAN and CNN.

Abstract

A screening system for early-stage oral cancer and precancerous lesions should be established because it is difficult to detect them even for specialists and they are often detected too late. In this paper, we propose a method for automatically classifying fluorescence images acquired by ALA-PDD (Photodynamic Diagnosis using 5-Aminolevulinic Acid) into three classes: Normal, Low-Risk, High-Risk. We augment a small image dataset by training GAN (Generative adversarial networks) with Differentiable Augmentation, and then train CNN (Convolutional Neural Network) for the classification by the augmented dataset. Experimental results show good classification results, which suggest that the combination of ALA-PDD and CNN classification is a promising method for oral cancer screening. Clinical Relevance- The method proposed in this paper has a potential to be used as a screening method for early-stage oral cancer and precancerous lesions, that is non-invasive, accurate, easy to use, and does not require specialization.