Automating chest radiograph imaging quality control.

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Diagnostic chest radiograph quality control is crucial for accurate interpretation.
• Manual quality assessment is time-consuming and prone to variability.
• Automating this process can enhance efficiency and consistency in radiological departments.

Purpose of the Study:

• To develop and evaluate convolutional neural network (CNN) models for automating chest radiograph quality control.
• The specific quality metrics targeted include lung inclusion, patient rotation, and inspiration level.
• To compare the performance of CNN models against human observer variability.

Main Methods:

• Trained six CNN models on 2589 posteroanterior chest radiographs.
• Utilized data augmentation techniques such as cropping and horizontal flipping.
• Preprocessed images by histogram equalization and resizing to 512x512 resolution.
• Investigated inter-observer variability in manual image annotation.

Main Results:

• CNN models achieved areas under the receiver operating characteristic curve >0.88 for lung inclusion, >0.70 for rotation, and >0.79 for inspiration.
• High inter-observer agreement was observed for manual annotations (e.g., 92% for left lung inclusion, 78% for rotation).
• A correlation was found between higher inter-observer agreement and smaller variance in network confidence.

Conclusions:

• Developed CNN models offer automated tools for effective quality control in diagnostic radiology.
• These models can provide immediate feedback on chest radiograph image quality.
• The automated system can serve as a valuable educational tool for radiologists and technicians.