Explainable Artificial Intelligence for Sickle Cell Anemia Prediction using Deep Neural Techniques
Sickle Cell Disease (SCD) presents substantial diagnostic challenges, and the integration of artificial intelligence (AI) into haematological analysis has the potential to improve early detection and clinical decision-making. However, traditional deep learning models lack transparency, limiting their acceptance in clinical applications. This study aims to develop a transparent and interpretable predictive framework by training a Multilayer Perceptron (MLP) for SCD classification and using an Explainable Boosting Machine (EBM), an inherently interpretable model, to provide post‑hoc and global interpretability of the underlying data patterns. This work separates the predictive MLP model from the interpretable EBM model, using the EBM to examine feature importance, trends, and clinical consistency, this method will provide better insight to the predictions made by the MLP Model. A dataset of 50 patient records with 12 haematological features was analysed. The MLP achieved strong predictive performance, and EBM provided clinically aligned explanations, with haemoglobin related biomarkers contributing most strongly to SCD predictions. Evaluation was done, using a confusion matrix and standard performance metrics showed high diagnostic utility while revealing dependent patterns that require clinical caution. The results demonstrate that interpretable models such as EBM can coexist with deep learning to support clinically trustworthy AI. Future work should focus on larger datasets, rigorous external validation, and comparison of emerging XAI techniques to improve diagnostic safety and clinician confidence.
Download Paper