A hybrid convolutional neural network-extreme learning machine with augmented dataset for dna damage classification using comet assay from buccal mucosa sample

Abstract

DNA is the information carrier in cells that are susceptible to damage, either naturally or due to external influences. Comet assays are often used by experts to determine the level of damage. However, the comet assays gathered with swab technique (Buccal Mucosa for example) often produced a higher noise level compared to ones that are cell-cultured, thus, making the analysis process more difficult. In this research, we proposed a novel way to assess the degree of damage from Buccal Mucosa comet assays using a hybrid of Convolutional Neural Network (CNN) and Extreme Learning Machine (ELM). The CNN was used to capture and extract spatial relation from every comet, while the ELM was used as a classifier that can minimize the risk of vanishing gradient. Our hybrid CNN-ELM model scored 96.96% for accuracy, while the VGG16-ELM scored 88.4% and ResNet50-ELM 76.8%.