Abnormality Detection in Smartphone-Captured Chest Radiograph Using Multi-pretrained Models

Lungs are the most important organs for living organisms. The Forum of International Respiratory Societies (FIRSs) told that most people are ignorant about lung disease, which kills more people than any other disease worldwide. According to Journal of community Hospital Internal Medicine perspectives, three of every 107 were clearly attributed to delayed diagnosis and treatment. There is a need for immediate or wet reading of the CXR reports as these pulmonary diseases are more dangerous that leads to death. Image-based telephone consultations using smartphones are becoming more and more popular. Integrating the algorithms of automated radiology report recognition CXR into smartphones offers many advantages. First, it gives medical professionals who need help interpreting CXRs or obtaining a second opinion access to radiologist-level expertise anytime, anywhere. Second, there is an opportunity for quality assurance as the algorithms are continually evaluated and adjusted for radiologists. We developed a transfer learning model using a modified CNN for sub-classifying various lung diseases, achieving high accuracies in the process, Abnormality Yes or No 92.18%, Enlarged Cardiomediastinum 86.67%, Cardiomegaly 80.19%, Lung Opacity 55.53%, Lung Lesion 93.68%, Edema 72.15%, Consolidation 90.79%, Pneumonia 95.18%, Atelectasis 82.88%, Pneumothorax 87.96%, Pleural Effusion 62.75%, Pleural Other 96.27%, Fracture 93.91%, Support Devices 61.40%.