Designing a Non-Invasive Anemia Detection System for Sustainable Healthcare

Abstract

Technology is revolutionizing healthcare by improving non-invasive medical techniques for diagnosing diseases, as disease images are crucial for numerous medical diagnoses. One of the common illnesses that affects people's health worldwide, particularly children and women of childbearing age, is anemia. By using cutting-edge technology to handle this problem, the prevalence will be significantly lower. A disorder known as anemia occurs when the blood's hemoglobin content falls below normal. In this work, we created a system to find people who are anemic using Mobile Net’s deep learning models, versions 2, 3 small, and 3 large. These models were trained and tested on created a dataset of 10,636 color palm images of adults that were labeled as either anemic or not anemic. The high training was Mobile Net v2 accuracy 99.9%, while v3s 81%, v3L 73.7%, and the best test results were in Mobile Net v2 (accuracy 95.77%, precision 96.05%, recall 95.51%, f1-score 95.78%). We have created models to evaluate medical images, automate estimations, reduce diagnosis time and error, and lower fatality rates. They support the sustainable development objectives, especially SDG 3, by encouraging early intervention for better patient quality of life.