Hybrid Feature Extraction and Classification for Conjunctiva-based Anemia Screening

View at publisher Summarise with VIRAT
Authors: Gourishetty Sindhusha, Rupesh Kumar Mishra, R Jegadeesan
Cited by: 0
References: 18

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Anemia is still a significant worldwide health problem, especially in low-resource environments with restricted availability of laboratory-based blood assays. This study proposes a non-invasive, image-based anemia screening system with conjunctiva images and ensemble deep learning. The proposed method combines three pre-trained CNNs, namely VGG16, ResNet50, and InceptionV3, for capturing complementary deep features of conjunctival images. These features are combined and optimized by using PCA and classified through a fully connected neural network. The performance of the model was assessed using an augmented and balanced dataset of 2,000 conjunctiva images from the EYES-DEFY-ANEMIA dataset. Experimental results show that the proposed ensemble model has advantages compared with individual CNN architectures, and reaches an accuracy of 96.8%, a precision of 96.9%, a recall of 99.8%, an F1-score of 98.3%, and an AUC measure up to be 98.4%. Results demonstrate that the deep feature fusion with dimensionality reduction is effective for robust and non-invasive anemia screening, suggesting its potential use in point-of-care and m-health settings.

Fields of Study & Subjects

Digital Imaging for Blood Diseases Artificial Intelligence in Healthcare Retinal Imaging and Analysis

Publication details

Journal / source
International Conference on Image Processing and Capsule Networks (ICIPCN)
Pages
674–678
Publication date
27 January 2026
Language
EN