Pneumonia detection in X-ray chest images based on convolutional neural networks and data augmentation methods
Jakub Garstka, Michał Strzelecki
Abstract
Artificial intelligence is gaining in importance in our everyday lives. Convolutional neural networks (CNN) are a very promising and perspective technology in the area of medical images processing, where it could contribute to diagnostics becoming easier and more reliable. Accurate diagnosis is an important factor in the selection of proper and effective treatment. In this paper, a self-constructed convolutional neural network trained on a relatively small dataset for classification of lung X-ray images is presented. This CNN enables classification into one of three categories: healthy, those with bacterial pneumonia, and those with viral pneumonia. Such classification, that considers pneumonia distinction, is rather uncommon among scientific publications. Also, a comparative analysis of the degree of impact of data augmentation on the model's performance and prevention of overfitting was performed. The obtained accuracy of the categorical classification has reached the level of 85% while the sensitivity was equal 0.95. Such results are promising for further work and improvement.