Development of Deep Learning-based Automatic Scan Range Setting Model for Lung Cancer Screening Low-dose CT Imaging

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Oncology

Background:

• Lung cancer screening with low-dose computed tomography (CT) is crucial for early detection.
• Accurate scan range setting is essential for effective screening and minimizing radiation exposure.
• Current methods for determining scan ranges may be time-consuming or prone to variability.

Purpose of the Study:

• To develop and evaluate a deep learning-based system for automatic detection of lung cancer screening scan ranges on low-dose CT.
• To compare the efficiency and accuracy of this automated system against radiographer performance.

Main Methods:

• A retrospective study utilized 1984 low-dose CT scans for analysis.
• A deep learning algorithm was developed using 1144 scans (915 training, 229 validation).
• The algorithm's performance was evaluated on a test set of 240 scans, comparing its output to actual lung boundaries and radiographer-defined ranges.

Main Results:

• The deep learning system demonstrated high accuracy in predicting scan ranges, with mean differences of 4.72 ± 3.15 mm for the upper boundary and 16.50 ± 14.06 mm for the lower boundary compared to actual lung boundaries.
• Accuracy for the upper boundary was 97.08% with 0% over-scanning.
• Accuracy for the lower boundary was 96.25% with 29.58% over-scanning.

Conclusions:

• The developed deep learning system effectively predicts lung cancer screening low-dose CT scan ranges.
• The system achieves high accuracy using only frontal scout images.
• This automated approach holds promise for enhancing the efficiency and consistency of lung cancer screening protocols.