A hybrid Fuzzy Logic and Convolution Neural Network (FIS-CNN) for automatic detection and classification of objects in comet assay images

Abstract

deep learning algorithms are able to discover many complex features in large datasets. The manually based feature extraction may lower the accuracy of the information in addition to wasting time. In specific, with the huge databases it becomes complicated to use the conventional feature extraction methods. Thus, researchers have tended to use convolutional neural networks to detect and classify objects in images instead of using traditional methods. Detection of DNA damage is one of the most important topics in this era because it contribute to diagnosing many diseases at an early stage, as well as knowing the stages of disease development by deciding the degree of damage to the DNA. This study suggests a hybrid Mamdani fuzzy logic (Type-2) for detecting edges of each object in images using the the (FIS-CNN) model. The proposed model is based on preprocessing image enhancement using adaptive histogram equalization and segmenting processing in morphology operations for each object in the images. Then patterns of comets are detected using the CNN network and classified into five scores automatically. The experimental results conducted on the database have achieved a high-performance precision of 94.34% . The proposed approach compared to similar modern methods with a competitive performance. In addition, the proposed approach can detect comets that are difficult to see with the human eye.