Automated Anemia Classification and Hemoglobin Level Prediction using Deep CNN and GLCM Features of Palpebral Conjunctiva Images

Anemia is a common medical condition affecting millions worldwide, particularly in developing countries. Early detection of anemia is crucial for prompt treatment and prevention of its potential complications. In recent years, deep learning (DL) has shown great potential in various medical applications, including medical image classification, anomaly detection, and segmentation. This study proposes a transfer learning-based approach using a pre-trained DL model to detect anemia from palpebral conjunctiva images. The proposed method utilizes a pre-trained DenseNet-201 model and fine-tuned it on a target dataset of palpebral conjunctiva images to detect anemia. Deep features of palpebral conjunctiva images computed from the fine-tuned DenseNet-201 are fed to MLP to identify anemia. The performance of the proposed method is evaluated on a publicly available anemia dataset, and the results show that the proposed method achieves an accuracy of 93.7 % in detecting anemia from palpebral conjunctiva images. In addition to anemia classification, we computed the hemoglobin level of palpebral conjunctiva images based on the gray-level co-occurrence matrix (GLCM) statistical properties. The statistical properties of GLCM are given to support vector and polynomial regressors, and the mean value of the predicted scores of both regressors is used to estimate the hemoglobin level. Experimental results show that the proposed model achieves an average root mean square error of 0.72 for conjunctiva images.