Deep Learning-Based Histological Scoring of Cerulein-Induced Acute Pancreatitis Rat Model

In an experimental rat model of acute cerulein-induced pancreatitis, we aimed to investigate the ability of the deep neural network-based program to distinguish damaged cell structures in histological preparations derived from rat pancreatic tissues. After the pancreatic tissues of all rats underwent the paraffin procedure, 3-4 pm thick sections were taken from the paraffin blocks, stained with hematoxylin-eosin dye, evaluated with a light microscope and photographed using a light microscope. 89 mixed-size microscopic images are resized at 224*224 diameter. The datasets were divided into train, validation and test groups. The algorithm used in this study was based on the NAS-Net-Mobile and ResNet-101 models from MATLAB Transfer Learning. By increasing the number of samples in the method we use in histology, both the evaluation performance and time consumption are reduced with the Al we use. The accuracy rate we obtained with NAS-Net mobile was determined to be higher than ResNet-101.