Deep Learning Model-Based Detection of Anemia from Conjunctiva Images

Abstract

Objectives: Anemia is characterized by a reduction in red blood cells, leading to insufficient levels of hemoglobin, the moleculeresponsible for carrying oxygen. The current standard method for diagnosing anemia involves analyzing blood samples,a process that is time-consuming and can cause discomfort to participants. This study offers a comprehensive analysisof non-invasive anemia detection using conjunctiva images processed through various machine learning and deep learningmodels. The focus is on the palpebral conjunctiva, which is highly vascular and unaffected by melanin content. Methods:Conjunctiva images from both anemic and non-anemic participants were captured using a smartphone. A total of 764 conjunctivaimages were augmented to 4,315 images using the deep convolutional generative adversarial network model to preventoverfitting and enhance model robustness. These processed and augmented images were then utilized to train and testmultiple models, including statistical regression, machine learning algorithms, and deep learning frameworks. Results: Thestacking ensemble framework, which includes the models VGG16, ResNet-50, and InceptionV3, achieved a high area underthe curve score of 0.97. This score demonstrates the framework’s exceptional capability in detecting anemia through a noninvasiveapproach. Conclusions: This study introduces a noninvasive method for detecting anemia using conjunctiva imagesobtained with a smartphone and processed using advanced deep learning techniques.