Malaria Detection System 

Automated Detection of Malaria Parasites Using Deep Learning and Explainable AI (XAI)

Abstract: Malaria is a life-threatening infectious disease caused by Plasmodium parasites. The traditional diagnostic method, thin blood smear microscopy, relies on manual examination by expert pathologists, a process that is both time-consuming and prone to human error. This project has developed an automated diagnostic tool using Convolutional Neural Networks (CNN) to detect malaria parasites from microscopic images. Beyond simple classification, the model employs Grad-CAM (Gradient-weighted Class Activation Mapping) technology to visualize the cellular regions the AI focuses on, thereby meeting "Explainable AI" (XAI) standards.

Methodology & Technology:

1. Dataset: The model was trained using the official dataset provided by the National Institutes of Health (NIH), containing 27,558 segmented cell images. The dataset is balanced between Parasitized and Uninfected classes.

2. Model Architecture: A multi-layered CNN architecture was designed to learn spatial hierarchies of features (edges, textures, parasitic spots) within the images. The architecture comprises feature extraction layers followed by dense classification layers.

3. Explainability (Grad-CAM): To prevent the model from operating as a "Black Box," the Grad-CAM algorithm was integrated. This algorithm uses the gradients of the final convolutional layer to generate a heatmap, highlighting the pixels (typically the purple parasite spots) that led the model to predict an infection.

Conclusion: The developed tool has demonstrated high accuracy on laboratory test data and is designed as a clinical decision support system to assist pathologists in rapid diagnosis.