Mobile-Based Deep Convolutional Networks for Malaria Parasites Detection from Blood Cell Images
Nazmul Shahadat · 2023
This paper uses a novel mobile-embedded and parameter-efficient convolutional network architecture to detect malaria parasites, a life-threatening disease, from the microscopic blood smear images. Recent research on malaria parasite detection has used computer-aided 2D convolutional neural networks (CNNs) with and without transformers and attention-based networks. Many mobile-based network architectures have also been analyzed for the publicly available microscopic blood smear image dataset. Like other computer-aided CNNs and mobile-embedded network architectures, we introduce a novel but straightforward, parameter-efficient, mobile-supported, and deep-learning architecture on this malaria image dataset to detect malaria parasites. Here, we apply the Squeeze-and-Excitation block with a spatial 1D CNN layer. Our deep learning-based proposed model performs with a testing accuracy of 99.52% on detecting malaria parasites from blood cell microscopic image modified dataset. Extensive evaluation was tested on the original malaria image dataset, where our proposed model shows 96.78% testing performance. Our proposed model with 2.2M trainable parameters and 12.6M FLOPS achieves state-of-the-art accuracy for the original and modified malaria detection image datasets.