Improved deep convolutional neural networks using chimp optimization algorithm for Covid19 diagnosis from the X-ray...

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Computational Biology

Background:

• Deep Learning (DL) and Deep Convolutional Neural Networks (DCNNs) are crucial for accurate COVID-19 detection in radiological images.
• Gradient Descent-Based (GDB) algorithms, commonly used for DCNN training, face limitations such as parameter tuning, local minima, large dataset requirements, and lack of GPU parallelization.

Purpose of the Study:

• To develop a fast and accurate COVID-19 detector using DCNNs trained with the Chimp Optimization Algorithm (ChOA).
• To address the challenges of limited COVID-19 training data and the limitations of GDB algorithms.
• To enable parallel implementation of DCNN training using GPUs.

Main Methods:

• Proposed a novel DCNN architecture trained with ChOA (DCNN-ChOA) for fully connected layers.
• Utilized two public datasets (COVID-Xray-5k and COVIDetectioNet) for benchmarking.
• Compared DCNN-ChOA against standard DCNN, DCNN-GA, and MSAD, employing an ensemble of ten DCNN-ChOA models with weighted averaging based on validation accuracy.

Main Results:

• The proposed DCNN-ChOA achieved a validation accuracy of 99.11%, significantly higher than LeNet-5 DCNN's ensemble accuracy of 84.58%.
• The DCNN-ChOA detector demonstrated over 99.11% accuracy with a false alarm rate below 0.89%.
• Class Activation Maps (CAM) successfully identified COVID-19-infected regions consistent with clinical findings, validated by experts.

Conclusions:

• The DCNN-ChOA model offers superior performance compared to existing COVID-19 detection methods.
• The Chimp Optimization Algorithm effectively trains DCNNs for medical image analysis, especially with limited datasets.
• The developed model shows significant potential for rapid and reliable COVID-19 diagnosis in clinical settings.