Ensemble Deep Feature Fusion for Noninvasive Anemia Classification using Conjuctiva Images

As a Major public health issue, anemia affects more than two billion people around the globe. Many resource-constrained environments lack well-equipped labs, which are necessary for many conventional diagnostic techniques which depend on invasive blood testing. This paper presents a noninvasive, image-based diagnostic system using conjunctival images and deep learning methods to diagnose anemia in orderto overcome these constraints. Three well-known convolutional neural networks—VGG16, ResNet50, and InceptionV3—each adding distinct feature representations make up the proposed model. Principal Component Analysis (PCA) fuses and optimizes these elements; the resultant vector is categorized by a fully connected neural network. Using 2,000 high-resolution conjunctival images from the CP-AnemiC dataset, trained and evaluated, the ensemble model obtained a classification accuracy of 96.8%, well above single CNN architectures. This work emphasizes the possibility for non-invasive, AI- powered solutions in real-time anemia screening and diagnosis.