Antral Ulcer Detection Using Deep Learning

Abstract

An ulcer is a break on the skin, in the lining
of tissues, or on an organ. An antral ulcer is found in the
antrum part of the stomach. An antral ulcer can have many
complications, such as bleeding, perforation, obstruction, and
even cancer, if not treated properly on time.
In the remote part of Nepal, antral ulcer diagnosis is very
challenging due to the scarcity of trained gastroenterologists.
Most of the patients visit higher centers, which are located only in
urban cities, for proper management. Those patients who cannot
afford higher centers, they just take the medicine without proper
diagnosis. Those who visit the higher center present very late,
so there are more complications of disease. Among them, a few
complications lead to serious problems, even the death of the
patients.

This research study has been conducted to fulfill this huge
gap. Hence, AI-based applications help to eliminate delayed
management and early diagnosis and treatment of the patient. It
focuses on having a quick diagnosis of antral ulcer to have fast
treatment that prevents the serious or complex issue.
Various deep learning models based on Convolutional Neural
Networks (CNNs) are used, such as ResNet50, VGG16, Mo
bileNetV2, Inception-ResNet, and CNN. Here, a total of 967
images have been used to train the model, including 485 normal
and 482 ulcers. The performance of ResNet50 shows better as
compared to other applied models, having a validation accuracy
of 96 percent, with precision, recall, and F1-scores all above 96
percent.