Deep Learning Model for Disease Prediction Using Gastrointestinal-Endoscopic Images

Abstract

Modern-day science has helped in the cure of such anomalies and with the advancement in computer aided treatment, we can detect such anomalies at an earlier stage. This paper is based on a deep learning-based approach for early detection of disease, using GI-Endoscopic images. The work done is using the readily available KVASIR data set containing 8000 images categorized into a structured group of 8. In order to reduce the noise, we manually treated the data set beforehand using equal amounts of masking techniques in each category. In addition, we expanded the data set to increase accuracy, we used the previously accessible models, we increased parameters by adding layers, and we applied transfer learning to get a higher accuracy compared to the prior models. In order to speed up calculation and increase accuracy, we can also apply this model to WCE procedures by running the data set through it. Our model employs a very contemporary technique to categories, enhancing the model to live frames that are categorized as healthy or unhealthy achieving 96.89% accuracy.