Quoc Khanh Tran

Quoc Khanh Tran

B.Sc. student in Data Science and AI

Multi-Domain Learning for Plasmodium Development Classification

[Thesis] [Presentation Slide] [Github Code]

Context of the Research Project

The research project was conducted at Hanoi University of Science and Technology (HUST), Vietnam, from September 2024 to January 2025. It was part of a research project collaborated with French Armed Forces Center for Epidemiology and Public Health (CESPA) and French Armed Forces Biomedical Research Institute (IRBA) on developing an automatic clinical testing for malaria medicine. Project advisors are Dr. Nguyen Thi Oanh at HUST, Assoc.Prof. Muriel Visani at La Rochelle University, and Asst.Prof. Thierry Urruty at Poiters University. The project also served as my Bachelor’s graduation thesis at HUST. The thesis achieved the highest score at Computer Vision Thesis Defense Committee, Winter 2024 Semester. We are in progress of publishing our work.

Summary

Malaria caused by plasmodium parasite is a fatal disease that burdens society with medical and socioeconomic issues [1]-[3]. To mitigate the impact of the disease, information about development stages of the parasite is important for clinical testing of new medicine. However, microscopic diagnosis is time-consuming and requires highly skilled microscopists [4]. To this end, the project aimed at developing a deep learning model to automatically classify parasite into life cycle development stages. Specifically, multiple datasets were employed to enrich minor classes data to tackle data imbalance issue of the plasmodium life cycle development classification. Multi-Domain Information Fusion module was proposed to bridge the domain gap. Quantitative, qualitative evaluation, and analyses were carried out to show the effectiveness of our proposed methodology.

Contributions

My contributions to the project are summarized as follows:

  • Conducting exploratory research and literature review on plasmodium classification and detection.
  • Defining the research gap in addressing data imbalance in classification task.
  • Surveying and investigating multi-domain learning, domain adaptation methods for application to the problem.
  • Designing multi-domain fusion methodology and evaluation strategy.
  • Conducting easibility study of implementation and computational resource requirements.
  • Implementing the proposed methodology with Python deep learning frameworks such as PyTorch, MMPreTrain, PyTorch Geometry, Numpy.
  • Conducting analyses of experiment results and iteractively making modifications.
  • Communicating rsearch findings through thesis report writing and thesis defense at HUST.

References

[1] World Health Organization, Malaria, https://www.who.int/newsroom/fact-sheets/detail/malaria, Accessed: December 28, 2024, 2024.
[2] J. Sachs and P. Malaney, “The economic and social burden of malaria,” Nature, vol. 415, no. 6872, pp. 680 685, 2002.
[3] K. E. Halliday, S. S. Witek-McManus, C. Opondo, et al., “Impact of school-based malaria case management on school attendance, health and education outcomes: A cluster randomised trial in southern Malawi,” BMJ global health, vol. 5, no. 1, e001666, 2020.
[4] Z. Jan, A. Khan, M. Sajjad, K. Muhammad, S. Rho, and I. Mehmood, “A review on automated diagnosis of malaria parasite in microscopic blood smears images,” Multimedia Tools and Applications, vol. 77, pp. 9801–9826, 2018.