ABSTRACT
The burden of oral cancer on the whole world in general and India in specific is increasing in an exponential order. Oral cancer is cancer that originates in the mouth region. The gold standard for diagnosing oral cancer is histopathological analysis. Hematoxylin and Eosin is the most common stain used in pathology labs as it helps in the staging and classification of most biopsy tissues. As this analysis is done manually, it allows errors in terms of misclassification. We have proposed a novel method to classify these Hematoxylin and Eosin stained photomicrographs through the fusion of the predictions of some of the most popular deep learning models. The deep learning models used in our experimental analysis include the winners and runners up of the Imagenet Large Scale Visual Recognition Challenge. Authors have achieved a successful classification of the photomicrographs into benign and malignant classes with the accuracy of 100% on the data consisting of 21 benign and 98 malignant images of 100x magnification.
