Developed in recent years, deep neural network
becomes the best method for rapid analysis of advanced
features and automation in medical image analysis. As a
second clinical opinion provided by artificial intelligence (AI),
it can reduce the physician's workload and reduce
misjudgment. This study collected 365,892 chest X-ray images
and clinical diagnosis reports through retrospective analysis,
and compared five different input image sizes and images that
generated by clinical labeling pre-processing in the
classification model building and testing. An AI trained chest
X-ray abnormal interpretation model by using DesNet121
neural network gave a test accuracy of 0.875. Deep neural
network shows the potential of accountable methods to help
lung classifications for normal and abnormal screening in
clinics.
