Generalizable AI approach for detecting projection type and left-right reversal in chest X-rays

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Deep Learning for Medical Diagnosis

Background:

• Manual verification of chest X-ray images for orientation and reversal is challenging and time-consuming.
• Ensuring correct imaging direction (e.g., anteroposterior, posteroanterior) is crucial for accurate radiological interpretation.
• Existing methods for image verification lack automation and efficiency.

Purpose of the Study:

• To develop and evaluate an AI-based system for automated verification of chest X-ray imaging direction and consistency with examination orders.
• To classify chest X-ray images into four categories: anteroposterior (AP), posteroanterior (PA), flipped AP, and flipped PA.
• To assess the impact of different training datasets on the AI system's classification accuracy.

Main Methods:

• Development of a deep convolutional neural network (DCNN) for image classification.
• Training the DCNN on multiple publicly available chest X-ray datasets.
• Testing the DCNN on both internal and external datasets to evaluate generalization.
• Utilizing Grad-CAM for visualizing the network's decision-making process.

Main Results:

• The DCNN accurately classified imaging directions and detected image reversal.
• Classification accuracy was significantly influenced by the training dataset, with performance dropping on external data.
• Training on a mixed dataset improved accuracy but still showed a decrease when tested on the COVID-CXNet dataset (76.0%).
• Visualization highlighted key areas like cardiac silhouette and arm positioning influencing classification.

Conclusions:

• AI, specifically DCNNs, holds significant potential for automating the verification of imaging direction and positioning in chest X-rays.
• The performance of AI models is highly dependent on the characteristics of the training data.
• Fine-tuning AI models to local data characteristics is essential for achieving optimal and reliable performance in clinical settings.