Pneumonia Detection Using Gradient Boosting on Selected Features Extracted from DenseNet

Shukla S.; Arya K.V.; Garg B.; Lezama P. · Lecture Notes in Networks and Systems (2023) · DOI: 10.1007/978-981-99-1912-3_17

Resumen

A severe respiratory condition that affects our lungs is pneumonia. It restricts the patient’s oxygen intake and makes breathing difficult. Worldwide, pneumonia is a leading cause of death for children under the age of five. The World Health Organization lists pneumonia as one of the leading global causes of child fatalities. In our proposed model, the features are extracted from the chest radiograph images using a fine-tuned deep learning model, DenseNet-201. Then the extracted features from the DenseNet-201 undergo feature selection using a combination of the Binary Sine Cosine Algorithm(BSCA), which is paired with the Adaptive Beta Hill Climbing (A β HC) local search algorithm for selecting the optimal feature set from all the extracted features. The searching algorithm runs for multiple iterations and picks up the optimal feature set. Using the Gradient Boosting Classifier, the selected optimal features are then used to classify the images into two classes, “Pneumonia” and “Normal” Chest Radiograph images. The proposed model has an accuracy of 98.92% in predicting pneumonia using chest radiograph images. When compared with the previous research work conducted on the same dataset, it is clear that our model outperforms the current state-of-the-art model which had an accuracy of 98.81%.

Instituciones: Universidad Nacional Federico Villarreal