Lung disease classification using chest X ray image: An optimal ensemble of classification with hybrid training

Problem: Even in today's highly developed and contemporary world, lung diseases are still devastating. The chest x-ray (CXR) is the main sources of identifying lung diseaseand deep learningmodels extensively play a major role in identifying the diseases. Aim: CXR has been used in various studies to identify lung illnesses in recent years.WedevelopaLung Disease Classification Model with Butterfly customized BES -Ensemble Model termed LDC-BCBES. Method: Four processes, including preprocessing, segmentation, feature extraction, and classification, are included in this work. The input image is filtered using a new Weighted Median Filtering (WMF) during the preprocessing stage. The filtered image is subjected to a segmentation process, where the ROI and non-ROI regions of the image are separated by the U -Net segmentation. Consequently, the features are Resnet-based features, Loop features, and Improved Local Gabor Binary Path features areextracted in the step of feature extraction. Result: The accuracy of the LDC-BCBES is 93.72 %, which is comparatively greater than DBN is 82.47 %, DQN is 65.79 %, RNN is 83.14 %, LSTM is 87.83 %, SVM is 87.81 %, Deep Convolutional Neural Network and Ensemble Learning (DEEPCNN + EL) is 65.26 % and CNN is 67.65 %, respectively. Conclusion: The extracted features are subjected to the classification stage, where an ensemble of classifiers is used that includes the models are Deep Neural Network (DNN), Long Short Term Memory (LSTM), DQN, and Deep Belief Network (DBN). At this point, a new training algorithm termed as BCBES is put into practice by optimizing the weights of LSTM.