Hybrid COVID-19 segmentation and recognition framework (HMB-HCF) using deep learning and genetic algorithms.

Area of Science:

• Medical Imaging
• Computer Science
• Artificial Intelligence

Background:

• Manual COVID-19 diagnosis is time-consuming, necessitating automated solutions.
• The rapid escalation of COVID-19 to pandemic status highlighted the need for faster detection methods.
• Computer science offers potential for developing automatic diagnosis systems for rapid disease detection.

Purpose of the Study:

• To propose a hybrid framework (HMB-HCF) for automated COVID-19 detection using deep learning and other computational techniques.
• To develop a lung segmentation algorithm (HMB-LSAXI) for precise feature extraction from X-ray images.
• To optimize hyperparameters using a genetic algorithm (GA) for improved model performance.

Main Methods:

• A hybrid framework (HMB-HCF) integrating deep learning (DL), genetic algorithm (GA), weighted sum (WS), and majority voting.
• A hybrid convolutional neural network (CNN) architecture combining custom CNNs and transfer learning (TL) models (VGG16, VGG19, ResNet50, etc.).
• Utilized a unified X-ray dataset from 8 public sources, employing regularization, dropout, and data augmentation to prevent overfitting.

Main Results:

• The hybrid CNN architecture achieved state-of-the-art performance metrics.
• VGG16 reported a 100% WS metric, including 99.78% accuracy and 0.9996 AUC.
• The HMB-HCF framework demonstrated strong generalization capabilities, validated on 13 diverse public datasets.

Conclusions:

• The developed HMB-HCF framework offers a highly accurate and efficient approach for COVID-19 detection from X-ray images.
• The study validates the effectiveness of integrating DL, GA, and ensemble methods for medical image analysis.
• The framework's performance and generalization suggest its potential applicability in clinical settings for rapid diagnosis.