MalariaScope: leveraging explainable ai for enhanced malaria diagnosis through deep learning-based blood cell classification

This study presents a novel methodology to improve malaria diagnosis through the application of deep learning techniques, particularly using long short-term memory (LSTM) networks for the analysis of blood cell images acquired from a digital microscope or a smartphone. Our research aims primarily to make a substantial advancement in the accuracy of malaria diagnosis and detection. Indeed, we have leveraged the power of LSTM to identify complex sequential patterns in the data, which has enabled us to achieve a high accuracy rate exceeding 98% for malaria detection from red blood cells. We have also set ourselves another major objective, which is to improve the explainability of the results obtained as well as the reliability of our model. The reason we integrated the LIME framework (Local Interpretable Model-agnostic Explanations) is that it provides model-independent explanations for the predictions made by our LSTM network, thus addressing the growing demand for transparency in machine learning applications and facilitating a clearer understanding of the decision-making process. Nevertheless, the highlight of this work remains the implementation of MalariaScope, a web platform designed to provide users with unlimited online access and greater cost-effectiveness, integrating our LSTM-based classifier and, of course, the visualization of explainability, offering users the transparency and understanding of the decision-making process that is so lacking in AI. Our research, along with the methodology we used, has allowed us to ensure that the model we developed is not only accurate but also explainable and accessible, thus meeting the clinical needs for automatic Malaria diagnosis.