Machine learning approaches classify clinical malaria outcomes based on haematological parameters.

Area of Science:

• Medical Diagnostics
• Computational Biology
• Infectious Diseases

Background:

• Malaria remains a significant global health issue, affecting over 3.2 billion people.
• Distinguishing malaria, particularly uncomplicated malaria (UM), from non-malarial infections (nMI) is challenging.
• Limitations of rapid diagnostic tests (RDTs) necessitate alternative diagnostic support methods.

Purpose of the Study:

• To evaluate machine learning (ML) approaches for classifying non-malarial infections (nMI), uncomplicated malaria (UM), and severe malaria (SM).
• To utilize haematological parameters for accurate malaria diagnosis via ML.
• To develop a precision medicine tool for malaria classification.

Main Methods:

• Collected haematological data from 2,207 participants in Ghana (nMI, SM, UM).
• Tested six ML approaches, employing an artificial neural network (ANN) for multi-classification.
• Developed binary classifiers with LIME for distinguishing malaria types from nMI.

Main Results:

• The multi-classification model achieved >85% accuracy in distinguishing malaria from nMI.
• Platelet, RBC, and lymphocyte counts were key classifiers for UM (0.801 test accuracy).
• Mean platelet volume and mean cell volume uniquely classified SM (0.96 test accuracy).

Conclusions:

• ML approaches offer a feasible tool for clinical decision support in malaria diagnosis.
• This study provides proof-of-concept for ML in classifying UM and SM from nMI.
• Future integration of ML into clinical algorithms can enhance diagnosis and treatment monitoring for febrile illnesses.