查看文章

Anemia is a worldwide health issue. To diagnose anemia, blood must be drawn to examine the hemoglobin level. The procedure is time-consuming and labor-intensive. The existing Artificial Intelligence (AI)-based anemia detection methods in literature have shortcomings, including, i) specially designed data collection device, ii) manual feature extraction, iii) small data size for training the model, and iv)user’s trust in AI prediction. In this paper, we aim to provide a non-invasive model of anemia detection from visible signs. We trained a CNN model on eye-membrane image data collected from real patients and open image sources. Our model predicts anemic patients with good accuracy at 98%. In addition, we proposed the explainable AI method as a part of the non-invasive diagnosis to enhance the user’s trust in the CNN model’s prediction.