Revolutionizing Malaria Diagnostics: EfficientNetB3-Based Analysis of Parasitized and Uninfected Cells
Shiva Mehta, Danish Kundra · 2025
Malaria still affects millions of people all over the world and is a serious disease, the diagnosis of which is important for success treatment. Mohammed Ahmed et Al working on EfficientNetB3: A Convolutional Neural Network for Automated Detection of Malaria using Microscopic Cell Images. The NIH Malaria dataset was employed which comprised 27,558 images divided into training set (24,000 images), validation set (3,600 images)and a test set (2,400 images). Classification accuracy was recorded at 96.3% for EfficientNetB3 outcompeting VGG16 (93.1%) as well as ResNet50 (94.7%). We saw that the observational model had great recall of 96.7%, good precision of 95.8% as well as high F1-score of 96.2% for a balanced detection of parasitized and uninfected cells. It also yielded an area under the curve (AUC) of 0.98 for EfficientNetB3 superior discrimination of classes. Besides achieving a high level of accuracy, EfficientNetB3 was learned with only 12 million parameters, while VGG16 consists of 138 million parameters and ResNet 50 – 25,6 million parameters. Due to this reduced complexity, inference was done in 0.8s with a batch size of 32, compared to the 1.3s with ResNet50 and 3.2s with VGG16 in the same instance. Consequently, the study indicates that EfficientNetB3 holds promising possibilities for being used in day-to-day practicality in low-resource contexts because it is both diagnostically accurate and computationally efficient. The work in the future will concentrate on the evaluation of the model’s abilities to be general on different datasets and real-challenges of clinical applications to improve its availability and effectiveness for malaria diagnostics worldwide.