Implementation of Intelligent Pneumonia Detection Model, Using Convolutional Neural Network (CNN) and InceptionV4 Transfer Learning Fine Tuning

Abstract

In Pneumonia is a type of contagious lung infection that has caused many human deaths in the form of inflammation of the alveoli. Based on WHO data, pneumonia is a type of acute infection that has caused more than 450 million cases and 4 million deaths each year. Covid-19 is one of the global pandemics that triggered many pneumonia incidents. Chest X-rays (CXR) are an important part of patient care. Radiologists can use CXR features to determine the type of pneumonia and the underlying pathogenesis. Machine learning and deep learning technologies are used to automatically detect various human diseases, thus ensuring smart healthcare. CXR features are more suitable to be analyzed by convolutional neural network (CNN). This algorithm is one of the typical deep learning architectures that has strong characteristics that are widely applied in the healthcare field. This study aims to develop a deep learning-based paradigm to distinguish Covid-19 patients from healthy and normal individuals by analyzing the presence of pneumonia disease symptoms on the CXR. This research provides an approach to the use of InceptionV4 transfer learning type in performing classification on CXR images. There are three main approaches carried out, namely making a standard CNN model, optimizing transfer learning xceptiion and fine tuning. The performance metrics results show a recall value close to 100% with a model accuracy value of 88%. Achieving a high enough recall value with a relatively small dataset makes the model built is considered to have good capabilities. The ability is also confirmed by the high ROC-AUC value with a value of 0.965