Automated matching of temporally sequential CT sections

Area of Science:

• Medical Imaging
• Radiology
• Computational Anatomy

Background:

• Evaluating patient response to therapy using thoracic computed tomography (CT) scans necessitates matching anatomically equivalent sections across sequential scans.
• Manual selection of these sections is time-consuming and subjective, potentially impacting treatment assessment accuracy.

Purpose of the Study:

• To develop and validate an automated method for selecting anatomically equivalent sections in temporally sequential thoracic CT scans.
• To expedite the process of matching baseline and follow-up CT scan sections for improved therapy response evaluation.

Main Methods:

• An automated method based on normalized mutual information (NMI) was developed to compare sections from baseline and follow-up CT scans.
• Sections were matched by translating and rotating follow-up sections against a baseline section, with NMI quantifying similarity.
• The method was tested with and without a thoracic segmentation pre-processing step on 22 patient scan pairs.

Main Results:

• The automated method without segmentation achieved an 81.8% accuracy in matching sections within the range selected by human observers.
• Implementing thoracic segmentation as a pre-processing step improved the accuracy of the automated method by 11%.

Conclusions:

• The developed automated NMI-based method is effective for selecting anatomically equivalent sections in thoracic CT scans.
• Thoracic segmentation enhances the accuracy of this automated matching process, offering a valuable tool for clinical practice.