Novel Crow Swarm Optimization Algorithm and Selection Approach for Optimal Deep Learning COVID-19 Diagnostic Model.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Biology

Background:

• The COVID-19 pandemic spurred numerous studies on computerized diagnostic models.
• The proliferation of diverse models necessitates a robust selection scheme for healthcare decision-makers.
• Evaluating deep learning model performance requires standardized criteria.

Purpose of the Study:

• To propose an integrated method for selecting optimal deep learning models for COVID-19 diagnosis.
• To employ a novel crow swarm optimization algorithm for evaluating and selecting models.
• To provide a tool for healthcare managers to assess COVID-19 diagnostic models.

Main Methods:

• Utilized two datasets of computed tomography (CT) images: 746 images (349 COVID-19 positive, 397 healthy) and 632 COVID-19 positive images.
• Applied 15 trained and pretrained deep learning models, including ResNet50, VGG16, and InceptionV3.
• Modified a crow swarm optimization algorithm with a fitness function to evaluate model performance based on nine metrics.

Main Results:

• ResNet50 achieved 91.46% accuracy and 90.49% F1-score on the first dataset, selected as optimal (fitness: 5715.988).
• VGG16 was optimal for the second dataset (fitness: 5758.791).
• InceptionV3 demonstrated the lowest performance across both datasets.

Conclusions:

• The proposed crow swarm optimization-based method effectively selects optimal deep learning models for COVID-19 diagnosis using CT images.
• ResNet50 and VGG16 are identified as high-performing models for specific datasets.
• This methodology serves as a valuable tool for healthcare managers in model selection and evaluation.