Non-invasive Anemia Detection from Conjunctival Images
Document Type
Conference Proceeding
Publication Title
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Abstract
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.
First Page
189
Last Page
201
DOI
10.1007/978-3-031-22061-6_14
Publication Date
12-14-2022
Keywords
AI, Anemia, CNN, Computer vision, Deep learning, Explainable AI, Non-invasive
Recommended Citation
R. Ferdousi et al., "Non-invasive Anemia Detection from Conjunctival Images," Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 13497 LNCS, pp. 189 - 201, Dec 2022. doi: 10.1007/978-3-031-22061-6_14
Additional Links
https://doi.org/10.1007/978-3-031-22061-6_14
Comments
IR conditions: non-described