PalmSense: Real-Time Anemia Detection Using Deep Learning on Palm Images

Anaemia does pose a challenge to the health sector of the world, especially in resource poor settings where most diagnostic techniques are invasive, expensive, and inaccessible. This study aims to develop a very accurate, non-invasive anaemia detection solution using deep learning models on palm images. Such tests, like blood tests, cannot be implemented in many lowresource settings. It is for this reason that the current research gap, particularly as regards the diagnostic capability of palm images using state-of-the-art deep learning methods, falls within the opportunity that may be contemplated here. This work used the ConvNext Tiny model, trained over a dataset of images taken from the palms of people in Ghana using transfer learning and heavy preprocessing of data, other CNN models including ResNet and VGG16. The results showed that the accuracy for ConvNext Tiny was 96.8% and rendered the highest data accuracy and also provided high speed compared with all other models. This approach promotes early detection of anemia, and not only is this a cost-effective method, but it is also accessible enough to be used in such sites of improving healthcare delivery where resources may be scarce.