Convolutional neural network evaluation of over-scanning in lung computed tomography

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Lung computed tomography (CT) scans often involve over-scanning in the Z-direction.
• This over-scanning leads to unnecessary patient exposure to ionizing radiation.

Purpose of the Study:

• To develop a convolutional neural network (CNN) for accurately determining Z-direction over-scanning in lung CT examinations.
• To assess the reliability and efficiency of CNN in quantifying over-scanning compared to traditional methods.

Main Methods:

• A CNN model was trained using 250 lung CT scans and validated on 100 scans.
• The CNN automatically segmented scans into cervical, lung, and abdominal areas, identifying over-scanning.
• Agreement between CNN and radiologist assessments was evaluated using kappa statistics.

Main Results:

• The CNN achieved 0.99 accuracy on the testing dataset.
• Excellent agreement (kappa=0.98) was observed between CNN and radiologist evaluations.
• Over-scanning was found to be approximately 22.6% across 1000 lung CT examinations.

Conclusions:

• Lung CT examinations frequently involve substantial over-estimation of the scanned area.
• This leads to unnecessary patient over-exposure to ionizing radiation.
• CNN provides an easy, reliable, and quick method for assessing and potentially reducing over-scanning in lung CT.