Automatic Measurement Application of Heart Area from Chest X-Ray Images Using the U-Net Deep Learning Method

Abstract

Heart health is a basic human right and a crucial component of global health justice. In an ever-more-advanced age, every task becomes simple due to science, technology, and information development. However, certain tasks are still performed manually. Therefore, innovation in computerized system design is required. The human heart area calculation was performed by combining image processing and deep learning techniques. Deep learning is a scientific subfield of machine learning, while image segmentation is a step in image processing. This study employed the U-Net segmentation method to identify different stages of heart area calculation. U-Net could conduct image segmentation with the small training dataset accurately. This study’s population consisted of 800 chest X-ray images obtained from the Kaggle website, with human hearts as the sample. The findings revealed that the training data with the U-Net architecture model acquired an accuracy of 09.98. However, the testing data accuracy was still determined manually. In this work, the U-Net model employed an input shape measuring 256x256, a kernel size of 3x3, and 50 epochs.