A Framework for Automatic Recognition of Cell Damage on Microscopic Images using Artificial Neural Networks

Despite several technological advances in the past years, the vast majority of microscopy examinations continue to be performed in a very laborious, time-consuming manner, requiring highly experienced personnel to spend several hours to visually examine each microscope slide. Due to recent improvements in modern Digital Image Processing, professionals that work on microscopic exams could benefit from new tools that can apply image processing possibilities to their specific field. We propose a framework consisting of an image segmentation stage, feature extraction, and then a Shallow Neural Network related to human perception. The framework is used to classify among 5 types of animal cell damage analyzed in a case study. The case study used applies the Single Cell Gel Electrophoresis assay (SCGE, also known as comet assay) to the cells of land mollusk Helix aspersa in order to measure the DNA damage caused by mutagenic agents. To train and analyze the performance of our approach, we used a dataset manually segmented by a biologist and comprised of 130 slide samples with labeled cells. Our framework proved to be robust, achieving an average accuracy of 88.3%.