Combination of the Features of Pre-trained Xception and VGG16 Models to Identify Childhood Pneumonia from Chest X-Ray Images

Pneumonia is a potentially fatal disease, especially for children, caused by lung infections from either bacteria or viruses. People who live in some impoverished countries, particularly infants, run a significant danger from it. The diagnosis of pneumonia can often be made through examination of chest X-ray data, but when the lungs have undergone surgery, fluid buildup, or got affected by lung cancer, it becomes difficult to diagnose pneumonia. In these cases, using Computer-Assisted Diagnosis can aid physicians in the diagnosis process. This study proposes a new deep learning model by combining the Xception and VGG16 models, two different transfer learning techniques, to detect pneumonia. To optimize the image data, this study employs image normalization and augmentation techniques as part of the image pre-processing. To create our suggested model, we first picked the two separate transfer learning models, Xception and VGG16, concatenated them, then flattened the concatenated model, added an extra LSTM layer, then added some additional dense layers, and lastly added a sigmoid layer. 5216 images were accumulated to prepare the model training in the “NORMAL” category and the “PNEUMONIA” category. From these two classes, a total of 624 images were used to test the results. The proposed model reached 92.31% accuracy, 93.5% precision, 90.5% recall, and 91.5% f1-score. The thorough exploratory study demonstrates that the suggested strategy works for a wide range of datasets.