Sickle Cell Diagnosis Using a Hybrid of Lightweight and Deep Hierarchical Neural Networks

Abstract

Sickle cell disease (SCD) continues to be a major health issue worldwide, especially in areas with limited resources like central India and Chhattisgarh, where affordable and infrastructure-independent methods for early diagnosis are not available. The research offers a hybrid deep learning model fusing DenseNet121 and EfficientNetB0 network architectures for automatic SCD determination from peripheral blood smear images. Through the use of feature-level fusion, the model gets the morphological feature extraction from DenseNet and the computational efficiency from EfficientNet, thus making a balanced trade-off between accuracy and resource feasibility. On an augmented erythrocyte dataset, the hybrid model was able to achieve better performance than the individual baseline networks and therefore, it reached 90% accuracy and an F1-score of 0.89. The study findings show that the proposed model is very robust in lowering false positives and negatives, which is very important for clinical reliability in low-resource environments. In addition to its technical value, the platform shows how AI can be a powerful tool for democratizing hematological diagnostics, providing a scalable, efficient, and accurate solution that promotes healthcare access fairness and paves the way for next AI-driven applications in computational hematology.