Performance of an automated registration-based method for longitudinal lesion matching and comparison to inter-reader...

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Accurate assessment of individual lesions in metastatic cancer patients is critical for treatment decisions.
• Longitudinal monitoring with medical imaging requires reliable lesion matching between scans.
• Inter-reader variability (IRV) in manual lesion matching can impact clinical decision-making.

Purpose of the Study:

• To evaluate the performance of an automated lesion-matching algorithm.
• To compare the algorithm's performance against the inter-reader variability (IRV) of human experts.
• To assess the efficiency of automated versus manual lesion matching.

Main Methods:

• Collected 40 pairs of longitudinal PET/CT and CT scans from patients with lung, head and neck cancers, lymphomas, and advanced cancers.
• Categorized cases by cancer burden: low (<10), intermediate (10-29), and high (30+ lesions).
• Two nuclear medicine physicians manually matched lesions, with consensus serving as the gold standard for comparison with the automated algorithm using precision, recall, and F1-score.

Main Results:

• The automated lesion-matching algorithm demonstrated performance statistically similar to IRV across all metrics and cancer cohorts.
• In high-burden cases, the F1-score for the automated method was 0.89 (vs. 0.93 for IRV), and in low-burden cases, it was 1.00 (vs. 1.00 for IRV).
• Automated matching was significantly more efficient, with a median time of 3.9 minutes compared to 30-60 minutes for human readers.

Conclusions:

• The automated lesion-matching algorithm successfully meets the performance benchmark set by inter-reader variability.
• This automated approach significantly expedites the lesion-matching process.
• Automated lesion matching enhances the consistency of longitudinal assessments in metastatic cancer patients.