Aggregated approach for interstitial lung diseases classification using attention based CNN and radial basis function neural network

The medical field has significantly advanced with advances in technology, with a focus on biomedical devices and early diagnosis. Image processing techniques and artificial intelligence are used to analyze the lung anatomy and ensure an accurate diagnosis of interstitial lung diseases. This study proposes an automated approach for identifying Interstitial Lung Diseases (ILD) using biomedical images. Computed Tomography (CT) biomedical images were used for analysis. This CT image was analyzed using both radiomic and deep learning features for efficient identification of ILD at an early stage. Here, radiomic features were extracted using gray-level properties and reduced using Particle Swarm optimization with inverse maximization of accuracy and precision as objective functions. The reduced features were then trained and tested using a radial basis function neural network (RBFNN). In parallel, an attention-based convolutional neural network was used to perform deep learning-based ILD classification using gray and local pattern images. Finally, both model outputs were aggregated for the final prediction by evaluating accuracy, precision, and F1-score. The proposed approach outperformed the ensemble approach for ILD classification by increasing its accuracy to 5 % for final prediction.