Explainable AI for early malaria detection using stacked-LSTM and attention mechanisms

Malaria remains a global public health challenge, affecting more than 247 million people and causing 619,000 deaths worldwide in 2024 (according to WHO). Rapid diagnosis is essential for effective treatment and to improve patients’ chances of survival. In this study, we propose an interpretable deep learning framework for accurate malaria diagnosis using blood smear images. Also, We evaluate and compare several baseline deep learning (DL) models (fundamentals), customized VGG-16 and VGG-19, as well as newer DL models such as Vision Transformer (ViT) and MobileNet, and, for the first time, a stacked long-short-term memory network (stacked-LSTM) with an attention mechanism for automatic detection of malaria from blood smear images. These models were trained and validated on a publicly available dataset of over 27.000 labeled blood smear images. The comparative and statistical study conducted in this research showed us that the proposed Stacked-LSTM model with attention mechanism outperformed all other approaches, achieving a classification accuracy (0.9912), sensitivity, specificity, precision, F1 score (0.9911), and area under the curve (AUC) superior to all other models. Despite their solid performance, these models are often considered ”black boxes” due to their lack of transparency in the decision-making process, which poses significant challenges in medical applications and fields where human life is at stake. To address this, we have integrated explainable AI (XAI) techniques, namely Grad-CAM and LIME, to improve the model’s interpretability. Our results demonstrate the complementary value of combining high-performance deep learning models with XAI methods to enhance trust and certainty in AI-assisted medical diagnosis, suggesting that our model can support early and interpretable malaria detection in clinical environments.