Comparison of Deep Learning Models for Efficient Classification of Gastric Abnormalities

Gastric cancer stands out as one of the most prevalent malignancies globally, holding the fourth position in terms of cancer-related fatalities. This intricate disease is shaped by a mix of environmental and genetic factors that play a role in its onset and advancement. In this study, 288 histopathological images comprising normal stage, stage_1 and stage_2 obtained from gastric tissue dataset are considered. Squeezenet and Inception V3 deep learning models are used to extract pixel level features. Efficiency of neural network in classifying the tissue changes is accessed by varing the activation and solvers. Results show that the features extracted using squeezenet and inception V3 models has the ability of differentiating tissue changes. to achieve an accuracy of 96in differentiating the tumor cells. High value of 99.1% for MCC clearly indicates that the features extracted using inception v3 features has the ability to identifying the tissue changes. ReLu-L-BFGS-B displays better peformance in identifying the tissue changes with an accuracy of 98.3% and 97.2% using inception v3 and squeezenet deep learning model. From the results, neural network model with logistic-adam and ReLu-L-BFGS-B is found to show comparatively high peformance in detecting the tissue changes in the images with accuracy and AUC values of 98.3% and 100% respectively.. From the study it is concluded that the inception v3 features has better ability than the features obtained from squeezenet model. This may be due to the overfitting of data which results as a results of increased feature length of squeezenet model compared to inception v3 model.