Deep Convolutional Spiking Neural Network optimized with Arithmetic optimization algorithm for lung disease detection using chest X-ray images

Lung disease is a most common disease all over the world. A numerous feature extraction with classification models were discussed previously about the lung disease, but those methods having high over fitting problem, consequently, decrease the accuracy of detection. To overwhelm this issue, a Deep Convolutional Spiking Neural Network optimized with Arithmetic Optimization Algorithm is proposed in this manuscript for Lung Disease Detection using Chest X-ray Images as COVID-19, normal and viral pneumonia. Initially, NIH chest X-ray image dataset is taken from Kaggle repository for detecting lung disease. Then, the chest X-ray images are pre-processed using the Anisotropic Diffusion Filter Based Unsharp Masking and crispening scheme for removing noise and enhancing the image quality. These pre-processed outputs are fed to feature extraction. In feature extraction process, the empirical wavelet transform method is used. These extracted features are given into Deep Convolutional Spiking Neural Network classifier (DCSNN) for detecting lung diseases. Here, the weight with bias parameter of DCSNN is enhanced based upon Arithmetic Optimization Algorithm (AOA), which improves detection accuracy. The simulation is executed in MATLAB. The proposed LDC-DCSNN-AOA technique attains higher accuracy, higher Precision, higher F-Score analyzed with the existing techniques, like Lung disease detection using Support Vector Machines optimized with Social Mimic Optimization (LDC-SVM-SMO), Lung disease detection using eXtreme Gradient Boosting optimized by particle swarm optimization (LDC-XGBoost-PSO), Lung disease detection using neuro-fuzzy classifier optimized with multi-objective genetic algorithm (LDC-NFC-MOGA), Lung disease detection using convolutional neural network optimized with Bayesian optimization LDC -CNN-BOA respectively.

Automated summary

A new method for lung disease detection is proposed in this paper, which improves the accuracy and performance by making improvements in pre-processing, feature extraction, and classifier optimization.