An adaptive filter for anemia screening using deep convolutional neural network
Jose Bernardo Lazaro, Jennifer C. Dela Cruz, Jocelyn F. Villaverde · Franklin Open · 2025
ABSTRACT Anemia remains a significant global health concern, affecting diverse populations and contributing to substantial morbidity. Early and efficient detection is crucial for timely intervention, yet conventional diagnostic methods often require laboratory access, limiting widespread screening capabilities, especially in resource-limited settings. This study introduces an automated anemia detection system powered by deep convolutional neural networks (DCNNs), CMOS image sensing, Adam optimizer, and multi-scale feature extraction (MSFE) to improve diagnostic precision and accessibility. Blood smear images from approximately 1,000 samples, sourced from rural clinics, community health programs, and mobile screening units, were preprocessed to enhance contrast and reduce noise. The DCNN model, incorporating multiple convolutional layers, adaptive filtering, and MSFE techniques, effectively distinguishes normal and anemic RBCs. The Adam optimizer facilitates stable model training, ensuring efficient convergence. MSFE enables comprehensive morphological, texture, and frequency-domain analysis, capturing RBC size variations, membrane irregularities, and hemoglobin-associated texture abnormalities. Performance evaluation metrics indicate an accuracy of 95.27%, with a 100% training-to-validation ratio, demonstrating high classification reliability. The precision, recall, specificity, and F1-score further validate the model’s robustness in anemia screening. Findings highlight that MSFE significantly enhances diagnostic accuracy, particularly in detecting microcytosis, macrocytosis, and poikilocytosis, along with texture variations linked to hemoglobin deficiency. This automated system aligns with advancements in digital healthcare, telemedicine, and mobile diagnostics, enabling rapid and accessible anemia screening in regions with limited laboratory infrastructure. Future research will focus on expanding dataset diversity, refining feature extraction techniques, and validating clinical applications to enhance scalability and robustness for widespread deployment in healthcare environments.