Digital processing of cytogenetics images for classification and segmentation

The objective of this paper is the development of a bioinformatics tool that allows the detection of the DNA nucleoides (with heads and tails) through the recognition of fluorescent images, in order to its subsequent classification according to the comet assay technique. It is sought to automate and manage their storage, and to optimize analysis processes in a cytogenetic laboratory. For this purpose, possible methods are evaluated and the one selected combines mathematical and computer algorithms, neural networks and fuzzy logic. The developed method performs detection and segmentation in three steps: first, a pre-process and initial segmentation of the raw image is carried out; second, the fragments obtained are classified by means of a convolutional neural network (cnn) into three groups: head, tail and bottom; third, heads or cores and tails are measured, re-analyzed, and classified according to its proportional ratio. This detection, segmentation and classification process was verified by means of case study with cytogenetic images during the comet assay in the General Cytogenetic and Environmental Monitoring Laboratory of Subtropical Biology Institute, UNaM-IBS-CONICET. The results presented through the evaluation of the model gave an accuracy of 0,92 and a Fmeasure of 0,96. These models combine classification methods that make use of hybrid algorithms, neuronal networks, neuro-fuzzy algorithms and classification functions.