A comparison of hybrid deep learning models for pneumonia diagnosis from chest radiograms

It is estimated that about 5–7 million children die from pneumonia every year. Pneumonia affects a large proportion of the world's population, approximately 7%, and is one of the most common respiratory illnesses in humans. Chest X-Ray is the most frequent way of diagnosing pneumonia. However, in certain countries, there is a scarcity of qualified radiologists, and the detection procedure performed by radiologists is not very accurate and quite difficult. The detection accuracy should be improved. So, this method is proposed to reduce the complexity of diagnosis and a helping hand of the radiologists. A Convolutional Neural Networking (CNN) architecture with 22 layers was created for this proposed method and three distinct machine learning techniques were used to extract and classify the CNN model's learned features. Support Vector Machine, Random Forest Classifier, and K-Nearest Neighbor. To add variety to the current dataset, certain data augmentation techniques were utilized. Overfitting was avoided by using a regularizer in the first dense layer and also data augmentation is utilized. The dataset used was Mendeley Data v2 having a total of 5856 chest x-ray images in both Pneumonia and Normal Class. After the classification of the RF, KNN and SVM classifiers differently using CNN model trained features, the accuracy obtained respectively were 99.52%, 96.55% and 97.32%. This proposed CNN-RF hybrid method is comparable with other existing conventional methods having accuracy of 99.52% and the AUC score of 98.7%.