Deep learning approach for automated hMPV classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Virology

Background:

• Human metapneumovirus (hMPV) causes significant respiratory illness, especially in vulnerable populations.
• hMPV diagnosis is challenging due to symptom overlap with other viruses and limited detection systems.
• Traditional methods lack speed and accuracy, particularly in low-resource settings.

Purpose of the Study:

• To develop a novel deep learning framework, hMPV-Net, for precise hMPV detection and classification.
• To address diagnostic challenges and improve accuracy in identifying hMPV infections.

Main Methods:

• Utilized Convolutional Neural Networks (CNNs) for binary classification of hMPV-positive and negative cases.
• Employed simulated image datasets for training and evaluation due to limited real-world data.
• Implemented data augmentation, weighted loss functions, and dropout regularization to handle dataset imbalance and improve robustness.

Main Results:

• hMPV-Net achieved 91.8% test accuracy, with precision, recall, and F1-scores around 92%.
• Demonstrated superior computational efficiency with only 3.2 GFLOPs, significantly less than ResNet-50 and VGG-16.
• The model effectively generalizes to clinical scenarios despite dataset imbalances.

Conclusions:

• hMPV-Net offers a highly accurate and computationally efficient solution for hMPV detection.
• The framework's efficiency makes it suitable for deployment in resource-constrained healthcare environments.
• This deep learning approach enhances the diagnostic capabilities for hMPV, improving patient care.