Tuberculosis (TB) and pneumonia are two commonly misdiagnosed respiratory conditions associated with severe complications and high mortality rates. Chest X-Rays (CXRs) are an inexpensive, expeditious method to identify such respiratory conditions. Thus, a model to distinguish between CXRs depicting lungs classified as normal, infected with pneumonia, or infected with tuberculosis would lead to faster and more accurate diagnoses. Our model was trained on a 10,500 images derived from a combination of five CXR datasets from the Guangzhou Women and Children’s Medical Center, Shenzhen Hospital, Montgomery County, Belarus, and ChestX-ray8. It reported a validation accuracy of 94.42% and has the potential to prevent widespread misdiagnosis.
Tuberculosis is commonly misdiagnosed as pneumonia because of their similar appearance in chest-xrays. This leads to delayed treatment and further complications from misdiagnosis. This model has the potential to aid radiologists in diagnosing patients by distinguishing between these two respiratory condtion. It will also help improve diagnosis time and enable doctors to analyze more chest x-rays, improving patient outcomes.
A 34-year old patient with symptoms of coughing, chest pain, and shortness of pain goes to an emergency and is ordered a chest-xray. Our model could be run on his chest-xray to provide an accurate diagnosis of whether he has pneumonia or tuberculosis.
PA (Posterior-Anterior) view chest X-rays
This model does not diagnose viral pneumonia, and the entirety of the set of images part of the pneumonia class used in training are chest x-rays of children.
Model performance metrics
Precision = 0.94
Recall = 0.93
F1 Score = 0.93
Accuracy = 0.9245
Specificity = 0.97
Aluna Research Institute
The Aluna Research Group is a student-led research group that aims to explore the intersection between medicine and AI by developing cutting-edge deep learning solutions to pressing healthcare issues.