Point of Care Noninvasive Screening Tool for Early Detection of Anemia using Smartphone

Hemoglobin, a crucial protein situated in red blood cells, plays a vital role in the identification and diagnosis of various conditions, including Anemia, through blood examinations. Anemia is a prevalent ailment affecting billions of individuals globally. The conventional means of diagnosing anemia typically entail invasive blood tests, which can be inconvenient, expensive, and challenging to access, especially in remote or underserved areas. Therefore, the innovative project aims to create a noninvasive point-of-care screening tool for early anemia detection utilizing a smartphone. Smartphone-based tools offer widespread accessibility, enabling the prediction and monitoring of anemia for a broad population, including those in remote or underserved regions. This research focuses on the development of AI-enabled models for precise estimation of hemoglobin levels using noninvasive data captured by smartphones. The primary objective is to facilitate early detection and prediction of anemia, enabling timely intervention and treatment to enhance patient outcomes. The proposed system introduces a unique approach to predict hemoglobin levels noninvasively by analyzing the appearance of fingernails. Hemoglobin levels are forecasted by extracting RGB values from nail images, which are then applied to various regression models for hemoglobin level determination. The machine learning models employed encompass Random Forest Regression, K-Nearest Neighbors Regression, Bayesian Ridge Regression, Ridge Regression, and Multiple Linear Regressions. Among the diverse machine learning techniques, Ridge Regression exhibits a low root mean square error rate of 2.07 and a mean absolute error of 1.51. Additionally, the deep learning model EfficientNet demonstrates a reduced root mean square error rate of 0.6591 and a mean absolute error of 0.624. These outcomes highlight the efficacy of the proposed approach in predicting hemoglobin levels accurately.