Diagnosis of respiratory diseases from chest X-ray images using deep learning models

Respiratory diseases are important health problems that affect millions of people worldwide every year from past to present. Accurate and rapid diagnosis of these diseases plays a critical role in the initiation of treatment. In traditional methods, errors that may arise due to subjective interpretations of radiologists and long diagnostic times negatively affect the treatment processes of patients. Fast and accurate diagnosis is of critical importance to reduce the burden on healthcare systems, especially during pandemic periods. This study was carried out to automatically diagnose respiratory diseases from lung X-ray images using rapidly developing deep learning technologies. In the study, a deep learning model trained on the COVID-19 Radiography Database dataset was developed based on the EfficientNetB1 architecture. In the pre-processing stages of the data, image dimensioning, data augmentation techniques and labeling operations were performed. The EfficientNetB1 model learned the pre-trained weights from the ImageNet dataset by learning transfer. The last layer of the model was removed from the system, the classification layer was added and the model was adapted to the new dataset. In the training phase of the model, the Adamax optimization algorithm and the loss function 'categorical_crossentropy' and the accuracy metric were used to evalute the performance of the model. The developed model classified four different respiratory diseases such as normal (healthy), pneumonia, COVID-19 and lung opacity with a high accuracy rate of 98.56% with a training time of 23726 seconds. While the model achieved a high accuracy rate, especially compared to other multiple classification models, the fact that it achieved this with a fast training time showed that the developed model is more efficient than other models. These results show that the model is a strong candidate for automatic diagnosis systems in the field of medical imaging. The study can be considered as an important step towards improving the quality of healthcare services by revealing the potential of deep learning in early diagnosis of respiratory diseases.