Detection of Anemia Based on Conjunctival Images Using a Convolutional Neural Network (CNN) Method

Abstract

A hemoglobin level below 12 g/dL is the primary indicator of anemia, a condition commonly found in adolescent girls. Laboratory blood tests, as a conventional detection method, are invasive, time-consuming, and costly. This study developed a non-invasive classification system for anemia and non-anemia based on conjunctival images using a Convolutional Neural Network (CNN), implemented on a real-time website. A total of 433 conjunctival images were collected comprising 206 images of anemia and 227 of non-anemia sourced from smartphone cameras and the Kaggle dataset, divided in an 80:10:10 ratio for training, validation, and testing. Preprocessing included resizing to 150 150 pixels, augmentation (flip, rotation, zoom, translation, brightness), and pixel normalization. The CNN architecture consists of three convolutional layers (32, 64, and 128 filters), max pooling, dropout, and a fully connected layer with sigmoid activation, trained using the Adam optimizer and the binary cross-entropy loss function until the 43rd epoch. The model achieved an accuracy of 88.37%, precision of 0.89, recall of 0.88, and an F1-score of 0.88. The model was integrated with a Flask-based REST API and MediaPipe Face Landmarker to automatically detect the conjunctival Region of Interest (ROI) via camera or uploaded images, thereby potentially serving as a fast, practical, and easily accessible tool for the initial screening of anemia among adolescent girls in schools and primary health care facilities.