An Effective Framework for Identifying Pneumonia in Healthcare Using a Convolutional Neural Network

Abstract

Pneumonia is now a life-threatening respiratory illness that can affect the lungs. Mainly the aged and children suffer the most. If the right diagnosis is not made, it could be fatal. So early diagnosis is very much needed to save many human lives. For diagnosis purposes Medical imaging, such as a chest x-ray can be utilized effectively and skilled radiologists are needed for this. Due to the blurriness of X-ray images, proper diagnosis can be difficult and time-consuming, even for radiographers with experience. As human judgment is involved, a pneumonia diagnosis may be erroneous. Hence, a deep learning-based automated system can be used to assist the radiographer in taking decisions more precisely and accurately. There have been several existing methods available for diagnosing pneumonia but they have accuracy issues. In this paper, we seek to automate the process of identifying and categorizing cases of pneumonia from CXR images deploying deep CNN. A deep CNN model has been built from scratch which will automate the process and provide high diagnosis performance. After passing through multiple convolutional layers and corresponding max pooling layers, the information is then fed into the dense layers. Lastly, using the sigmoidal function, the classification is performed. The model's performance improves as it simultaneously gains training and reduces loss.