Deep Fractional Max Pooling Neural Network for COVID-19 Recognition

Area of Science:

• Artificial Intelligence
• Medical Diagnostics
• Deep Learning

Background:

• Coronavirus disease 2019 (COVID-19) presents a significant global health challenge.
• Accurate and efficient diagnostic tools are crucial for disease management and control.

Purpose of the Study:

• To introduce a novel deep fractional max pooling neural network (DFMPNN) for enhanced COVID-19 diagnosis.
• To evaluate the performance of DFMPNN against state-of-the-art models.

Main Methods:

• A 12-layer DFMPNN model utilizing fractional max-pooling (FMP) instead of traditional max pooling (MP) and average pooling (AP).
• Implementation of multiple-way data augmentation (DA) to mitigate overfitting.
• Application of model averaging (MA) to reduce predictive randomness.

Main Results:

• The DFMPNN model was evaluated on a dataset comprising COVID-19, community-acquired pneumonia, secondary pulmonary tuberculosis (SPT), and healthy controls (HC).
• Achieved a micro-averaged F1 (MAF) score of 95.88% across 10 test runs.
• Fractional max-pooling (FMP) demonstrated superior performance compared to MP, AP, and L2-norm pooling (L2P).

Conclusions:

• The proposed DFMPNN model exhibits superior diagnostic performance for COVID-19 compared to 10 existing state-of-the-art methods.
• Fractional max-pooling is an effective pooling strategy for improving deep learning model accuracy in medical image analysis.