Diagnosing malaria still relies heavily on microscope examination of blood smears — which works, but is slow, hard to scale, and only as good as the person doing it. This paper describes a deep learning pipeline for binary malaria cell screening built on a modified AlexNet CNN. We used the public NIH/Kaggle malaria cell-image dataset, resized everything to 100×100 RGB patches, applied live augmentation, and trained a regularized CNN with batch normalization and dropout. Beyond just reporting accuracy, we evaluated the model with confusion matrices, ROC-AUC, precision-recall curves, calibration plots, and Grad-CAM heatmaps. We also compare the approach against a recent two-stage YOLOv4/DenseNet-121 system. This model doesn't do whole-slide detection or species identification — it's a clean, reproducible infected-vs-uninfected screener that can be understood, explained, and built on.
Phanindra et al. (Thu,) studied this question.