Malaria parasite classification framework using a novel channel squeezed and boosted CNN.

Area of Science:

• Medical Imaging
• Computational Biology
• Parasitology

Background:

• Malaria is a severe, life-threatening parasitic infection transmitted by mosquitoes.
• Early and accurate diagnosis of malaria is crucial for effective treatment and disease control.
• Current diagnostic methods require timely identification of Plasmodium parasites in blood smears.

Purpose of the Study:

• To develop and validate a deep convolutional neural network (CNN) for automated detection of malaria parasite-infected red blood cells.
• To introduce a novel STM-SB-RENet architecture integrating split-transform-merge (STM) and channel squeezing-boosting (SB) for enhanced feature extraction.
• To improve the accuracy and efficiency of malaria diagnosis using digital image analysis of thin blood smears.

Main Methods:

• A novel STM-SB-RENet deep learning model was designed, incorporating modified STM and SB blocks with region and edge operations.
• Transfer learning (TL) and discrete wavelet transform (DWT) were utilized to enhance feature representation and capture subtle variations.
• The model was trained and validated on the National Institute of Health Malaria dataset using hold-out cross-validation.

Main Results:

• The proposed STM-SB-RENet achieved high performance metrics: 97.98% accuracy, 0.988 sensitivity, 0.980 F-score, and 0.996 AUC.
• The model demonstrated superior performance compared to training from scratch and existing TL-based fine-tuned methods.
• The architecture effectively captured fine-grained features indicative of malaria infection.

Conclusions:

• The STM-SB-RENet deep learning model shows significant potential for accurate and automated screening of malaria parasite infections.
• This technique can aid in the early detection of malaria, facilitating timely clinical intervention.
• The developed CNN architecture offers a robust solution for malaria diagnosis in resource-limited settings.