Comet assay classification for buccal Mucosas DNA damage measurement with super tiny dataset using transfer learning

Comet assay or single cell gel electrophoresis assay (SCGE) is a method which is frequently used to measure the damage of DNA. The results of comet assay is a set of comet images, then the comet images are classified to measure the level of DNA damage. Currently, there are several softwares for comet assay image analysis, both free and commercial. Commercial software is quite expensive, while free software is limited, especially for buccal mucosa cell and super tiny comet image dataset. In this research, we propose a classification model for comet assay with super tiny image dataset which is taken from buccal mucosa cells. We propose a transfer learning based convolutional neural network (CNN) model. We have compared the transfer learning model with CNN-support vector machine (SVM) and ordinary CNN. In our experiments, we use super tiny dataset consisting of 73 images. Our transfer learning model gives an accuracy 70.5, while CNN-SVM gives 62.3 and ordinary CNN gives 63.5. We also compare our transfer learning model with most frequently used, free comet assay analysis software, OpenComet. Open-Comet gives an accuracy 11.5. Our transfer learning model is promising for comet assay for buccal mucosa cell and super tiny dataset. © 2019 Elsevier B.V., All rights reserved.