AI Chest 4 All.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Radiology Informatics

Background:

• Significant radiologist staff shortages exist in rural Thailand.
• Accurate and efficient screening for critical diseases like heart disease, lung cancer, and tuberculosis is essential.
• Existing diagnostic tools may be limited in accessibility or speed.

Purpose of the Study:

• To develop and evaluate AIChest4All, a deep learning model for disease screening using chest X-rays.
• To assist radiologists in Thailand, particularly in underserved rural areas.
• To classify chest X-ray images into six distinct categories, including specific diseases and general abnormalities.

Main Methods:

• Utilized deep learning for chest X-ray image classification.
• Trained the model on diverse datasets: NIH (14 observations), Montgomery, Shenzhen, Udonthani Cancer Hospital, and National Chest Institute of Thailand.
• Processed a total of 201,527 chest X-ray images.

Main Results:

• Achieved high accuracy for key diseases: heart disease (94.11%), lung cancer (93.28%), and tuberculosis (92.32%).
• Demonstrated strong performance with sensitivity values: heart disease (90.07%), lung cancer (81.02%), and tuberculosis (82.33%).
• Reported excellent specificity values: heart disease (94.65%), lung cancer (94.04%), and tuberculosis (93.54%).

Conclusions:

• The AIChest4All model demonstrates sufficient accuracy, sensitivity, and specificity for clinical application.
• The AI tool is currently deployed in several Thai government hospitals, providing free screening services.
• AIChest4All effectively addresses the need for diagnostic support in areas with limited radiologist availability.