A deep neural architecture for SOTA pneumonia detection from chest X-rays

Pneumonia among children is a leading cause of death in India, and it gains a lot of researchers' attention to develop early detection tools. Due to a lack of the number of radiologists, especially in rural India, the traditional method of diagnosing pneumonia does not address the real-time issues related to early stages. This paper presents a deep learning model, NASNet (Neural Architecture Search Network), pre-trained on ImageNet to predict pneumonia very early stage through chest x-rays of patients. With 2.6 million trainable parameters, the proposed model can run even on a mobile phone with good precision, recall, and an F1 score to detect pneumonia. This approach thus proves to be significantly better than the current state-of-the-art models. It can also help trained radiologists to get a second opinion/ validation of pneumonia diagnosis.