PRISM: a novel deep learning framework with resilient fuzzy whale optimization for automated COVID-19 detection from chest X-rays

The 2019 coronavirus disease (COVID-19) is an infectious disease that has suddenly and unexpectedly spread worldwide. A significant challenge is to quickly determine who is infected with coronavirus. To find probable Coronavirus patients without being intrusive, the current aim leverages Artificial Intelligence (AI) techniques such as Deep Learning (DL). The expert optimization approach presented in this research improves the speed and achieves more accuracy than conventional models.

This paper introduces a novel disease diagnosis model, where the input images are initially subjected to a pre-processing. Secondly, an encoder-decoder architecture named Adaptive Two-stage Lung Analysis and Segmentation (ATLAS) is proposed to localize the lung region and ROI-focused feature refinement, a feature extraction module to refine the significant one through an enhanced Histogram of Oriented Gradients (HoG) and a feature map from Convolutional Neural Networks (CNNs) (Resnet-101, DenseNet-121, and MobileNetV3) to extract the significant features. Then, the extracted features are given to feature selection using the proposed Resilient Fuzzy Whale Optimization Algorithm (RFWOA). Finally, the selected features are subjected to classification using the proposed Pandemic Resilient Intelligent Screening Model (PRISM) which combines Vision Transformer (ViT) with convolutional layers, residual blocks, attention mechanisms, cross-attention mechanisms, position-aware attention and Fully Connected (FC) layers.

Chest X-Ray (CXR) images are employed for the experimentation. Eventually, classified results from developed PRISM attained 99.15% accuracy and are evaluated with a comparative study to verify the competence in COVID-19 detection.

The results confirm the performance of the proposed and can aid for pandemic scenarios.