CT scan range estimation using multiple body parts detection: let PACS learn the CT image content

Area of Science:

• Medical Imaging
• Computer Vision
• Radiology

Background:

• Accurate determination of computed tomography (CT) scan range is crucial for quantitative analysis and comparison.
• Current methods often rely on metadata, which may not always be available or consistent.
• Developing an image-based method offers a more robust and universally applicable solution.

Purpose of the Study:

• To develop an efficient computed tomography (CT) scan range estimation method based on image data analysis.
• To enable quantitative comparison of CT scan ranges between different studies.
• To create a method independent of metadata analysis.

Main Methods:

• Projecting 3D CT data to 2D coronal images using a ray casting-like process.
• Utilizing trained 2D body part classifiers to identify anatomical structures.
• Employing structure grouping and structural voting for accurate patient scale and position estimation.
• Normalizing scan range positions on a standard human figure (feet=0.0, head=1.0).

Main Results:

• Trained classifiers for 18 body parts using 184 CT scans.
• Tested on 136 heterogeneous CT scans, achieving mean absolute errors of 1.2% for start and 1.6% for end positions compared to human observers.
• Maximum errors were 3.5% and 5.4% for start and end positions, respectively.

Conclusions:

• A novel CT scan range estimation method was developed using multi-body part detection and relative structure positioning.
• The proposed method demonstrates promising accuracy in preliminary tests.
• This image-based approach facilitates quantitative comparison of CT scan ranges.