Classification of canine fibroma and fibrosarcoma histopathological images using convolutional neural networks

Abstract

Within the scope of the study, a high-performance convolutional network model that can classify canine fibroma and fibrosarcoma tumors based on 200 high resolution real histopathological microscope images has been developed. In order to determine the network performance, the well-known network models (VGG16, ResNET50, MobileNet-V2 and Inception-V3) were subjected to training and testing according to the same hardware and training criteria. While comparing the models, 13 different performance criteria were used and performance calculations were made for each model separately. The results obtained seem extremely satisfactory. Compared to its counterparts, the proposed network model (FibroNet) contains fewer trainable items, while achieving a much higher performance value and training time is shorter than others. Thanks to low prediction error rate achieved with FibroNET network using real data, it seems possible to develop an artificial intelligence-based reliable decision support system that will facilitate surgeons’ decision making in practice.