Impact of Data Augmentation on Skin Lesion Classification Using Deep Learning

Abstract

The known peculiar type of cancer type is melanoma. It arises as pigment and is hard to find in the initial stages. The persistence level is 99% when identified in the early stage. Classification and identification of malignant tumors in skin lesions are crucial. The main goal is to sort the lesion images to seven important classes and identify the cancerous and non-cancerous tumors at the earliest using deep learning techniques. The efficient way for deep learning outcomes is to use a large volume and high-quality training dataset. Existing datasets are effectively not sufficient for training the model. The techniques for data augmentation are effective ways to build highly accurate classifiers from insufficient data. The proposed methodology offered the effective strategy for diagnosing the malignant tumor is a CNN-based model. CNN is specifically used to recognize and classify images. The framework is trained with data that has been labeled with the appropriate class. A similar framework has been trained with augmented and non-augmented lesion images for knowing the malignant lesions. The results are compared to both original data and augmented data. The model evaluated, the accuracy occurred for augmented data is 97.86%.