Malaria Early Diagnosis Based on Transfer Learning and CNN Architecture
Zul Indra, Yessi Jusman, Elfizar · 2023
Malaria has spread worldwide since the early 20th century and causes nearly half a million deaths each year. Malaria is actually a curable and preventable disease if treatment initiatives are carried out early and effectively. Confirming the presence of the malaria parasite earlier would make treatment of the disease more effective in reducing mortality. Unfortunately, this disease is often ignored because it is considered the common cold and is only diagnosed when it has reached a critical phase. This research is expected to be an alternative for early diagnosis of malaria. This study aims to develop a computer-assisted disease diagnosis (CAD) system enriched with deep learning algorithms to help diagnose malaria. This CAD system has the potential to provide rapid, inexpensive and reliable diagnosis of malaria, avoiding common detection errors. In making CAD applications, this study applies the CNN algorithm which will undergo modification of the architecture. In addition to accelerating training, this study applies Transfer Learning with models that have been trained on ImageNet data. Based on the results obtained, this study succeeded in surpassing previous research benchmarks with an accuracy value above 97%.