A novel NASNet model with LIME explanability for lung disease classification

•To eliminate the unwanted noises in the images, pre-processing carried out using Normalization and Dynamic Fuzzy Histogram Equalization (DFHE).•To extract the relevant features in the image using Adaptive Attention Based Deep Neural Network architecture.•To select the particular features using Squid Game Optimizer Algorithm.•For classification, a proposed novel Dense Convolutional NASNet model (DenseConvNASNet) is used.•To explain the proposed classification model deeply, Local Interpretable Model Agnostic Explanation (LIME) method is used. In current days, some of the most frequent and deadly lung disorders are tuberculosis, COVID-19, and pneumonia. Many previous studies provided a variety of techniques for detecting particular diseases. People who receive a negative diagnosis for one disease may be easily affected by COVID-19. Many existing methods suggested new approaches to detect COVID-19, but they failed to identify, so in this study, an adaptive deep learning method is used to categorize COVID-19. Initially, the images are pre-processed using normalization and dynamic fuzzy histogram equalization (DFHE) to remove unwanted noises in the images. After pre-processing, feature extraction is done using an Adaptive Attention Based Deep Neural Network architecture. Then, the best features are selected using the Squid Game Optimizer Algorithm. The Modified Dense Convolutional Neural Architecture Search Network (NASNet) model (DenseConvNASNet) recognizes COVID-19, pneumonia, and tuberculosis. This classification method is used for CXR classification in the given dataset. Finally, the detailed explanation of classified output is carried out using Local Interpretable Model Agnostic Explanation (LIME), and the performance metrics show the accuracy of the proposed method is 98.55%, precision of 98.45%, recall of 98.2% and F measure of 98.3%.