Non-invasive anaemia detection based on palm pallor video using tree-structured 3D CNN and vision transformer models

Abstract

ABSTRACT Anaemia is a common disease that affects billions of people worldwide and is caused due to low blood haemoglobin level. According to WHO statistics, anaemia is the most prevalent in developing and underdeveloped countries. Conventional invasive methods are prohibitively expensive and difficult to administer globally, necessitating a non-invasive, low-cost, and user-friendly solution. This study aims to develop a non-invasive anaemia detection system by combining cutting-edge computational approaches with the age-old practice of estimating blood haemoglobin levels by observing pallor in the palm. The proposed system operates on the basis of inducing changes in palm pallor with appropriate pressure application and release, measuring the rate of colour changes, and performing time-domain analysis to correlate with blood haemoglobin concentration. The video of colour changes in the palm caused by a customised device is captured using a smartphone camera and processed and analysed using deep learning models based on tree-structured 3-Dimensional Convolutional Neural Network (3D CNN) and Vision Transformer (ViT) for accurate estimation of haemoglobin levels. The proposed system ensures a sensitivity, specificity, accuracy and RMSE of 96.87%, 90.90%, 94.44% and 0.495, respectively, while run on a dataset consisting of palm pallor video samples of 531 individuals.