An optimized transfer learning approach integrating deep convolutional feature extractors for malaria parasite...

Area of Science:

• Medical Diagnostics
• Artificial Intelligence
• Parasitology

Background:

• Malaria, a disease caused by Plasmodium parasites transmitted by Anopheles mosquitoes, poses a significant global health challenge.
• Current diagnostic methods, primarily manual microscopy, are labor-intensive and require skilled personnel, creating a bottleneck in high-prevalence regions.

Purpose of the Study:

• To develop and evaluate an automated malaria diagnosis system using ensemble learning.
• To enhance diagnostic accuracy and reliability compared to standalone models and manual microscopy.

Main Methods:

• An ensemble learning model was developed by integrating multiple transfer learning architectures: VGG16, ResNet50V2, DenseNet201, and VGG19.
• Data augmentation and pre-processing techniques were employed to improve model robustness.
• The ensemble model was fine-tuned for optimal hyperparameter performance.

Main Results:

• The ensemble model achieved a high test accuracy of 97.93%, with an F1-score and precision of 0.9793.
• This performance surpassed standalone models, including Custom CNN (97.20% accuracy) and VGG16 (97.65% accuracy).
• The model demonstrated high reliability in classifying parasitized and uninfected blood smears.

Conclusions:

• Ensemble learning integrating transfer learning models offers a powerful approach to improve malaria diagnostic accuracy.
• This automated solution is scalable and suitable for resource-limited settings, reducing reliance on manual microscopy.
• The proposed method represents a significant advancement in malaria detection technology.