DGPCNet: anemia detection from multimodal palm, fingernails, and eye conjunctiva images using deep global pyramid convolutional network.

BACKGROUND

Anemia is one of the most common hematological disorders causing fatigue, increased vulnerability to infection, low birth weight, and threatening the physical and mental development of children. Hence, the early identification reduces the risk of patients and facilitates early intervention. On the other hand, the blood collection of the adults is simple, and the results will be precise, and directly show the cause of anemia in adults. Noninvasive techniques for anemia diagnosis are specific for young children and pediatric patients improving the cost-effectiveness.

RESEARCH DESIGN AND

METHODS

Consequently, this research focus on anemia detection for children using the proposed Deep Global Pyramid Convolutional Network (DGPCNet) model. Specifically, the proposed approach involves anemia detection through three anatomical regions, such as palm, fingernails, and eye conjunctiva images. In the proposed method, the Grayscale-based Deep Statistical Structural Edge features (GDSSE) provide enriched feature representation by integrating multiple feature perspectives.

RESULTS

The experimental results report that the proposed method using fused features attains 97.66% accuracy, 97.19% sensitivity, 98.14% specificity, 97.90% precision, 0.018 FPR, and 0.028 FNR for 90% of training.

CONCLUSIONS

Moreover, applying fused features in the proposed DGPCNet enables hierarchy-based learning thereby significantly boosting the accuracy and reliability of detection.