Performance comparison of CNN architectures for detecting Malaria diseases
Abstract
World is facing an acute health crisis for the disease named malaria caused by the bite
of female mosquitoes of parasite named genus plasmodium. From different research
of all time it is clear that this disease is not confined within a certain specific region
or area rather this infection is common all over the world. Many researchers from
all over the globe discovered many processes or techniques to determine malaria
infection from host body. Malaria Detection consumes huge time to detect. This
study aims to determine the infected malaria cells using deep learning algorithms
as it is important part in this advanced technological era to determine objects.
This research used deep learning algorithms like VGG-16, VGG-19, VGG-16 binary,
VGG-19 binary, Alexnet, MobileNet, ResNet34, ResNet50 and CNN2D to determine
malaria infected cells from images. Thereby, also finds the comparative analysis
between these algorithms to determine the best accuracy giving algorithm. From
the study it is evident that algorithms named AlexNet, VGG-16, VGG-19, VGG-
16 binary, VGG-19 binary, MobileNet, CNN2D, ResNet34 and ResNet50 give an
accuracy of 94.84%, 92%, 92%, 97.4%, 96.53%, 95.42%, 96.91%, 97.06% and 85%
respectively. From the comparative analysis between these nine algorithms, this
study concludes to and ResNet34 with model accuracy 97.06% as the best accuracy
giving algorithm.