AI-Based Reference Charts Enhance Understanding of CT Imaging Volumes

A recent study published in Radiology: Artificial Intelligence has utilized artificial intelligence to transform standard CT data into reference charts for 104 anatomical structures. This research, led by Christian Wachinger, PhD, from the Technical University of Munich, suggests that CT scans offer more quantitative data than is typically used in clinical practice.

The study analyzed nearly 8 million volume measurements from CT images to establish reference benchmarks across various adult age groups. Dr. Wachinger noted that many anatomical volumes do not simply increase over time but instead may initially grow, stabilize, and subsequently shrink. For this reason, the research involved analyzing repeated CT scans from the same patients over follow-up periods of up to eight years.

Key findings of the study include substantial sex-based differences, with males showing larger and more variable volumes. Meanwhile, the use of contrast agents was found to significantly increase volume measurements in certain organs and vascular structures, which poses implications for scan interpretation protocols.

Traditionally, reference charts have been used to compare individual patients against population norms — a tool commonly used in pediatric growth assessments and MRI-based brain studies. This study fills a gap by providing such benchmarks for adult anatomical structures, leveraging clinical imaging archives.

Dr. Wachinger emphasized the potential of standardized clinical imaging data to enhance understanding of age-related anatomical changes. Despite advances in automated CT segmentation, reference charts needed to interpret such data are lacking. This research utilized flexible statistical modeling to generate these insights, considering both average and variable volumes across populations.

The potential clinical utility of these reference charts was also explored through the example of cardiomegaly, where patients were found to align with higher centile scores, enhancing the interpretation of raw anatomical data.

Experts like Yuan Chai, PhD, from Northeastern University in China, praised the study, highlighting its role as a conceptual milestone in radiology AI. Dr. Chai underscored the importance of distinguishing statistical norms from biologically healthy baselines, cautioning that data from clinical populations might not always represent healthfully.

The study's demographic scope was limited and based on clinical rather than health-reference cohorts, which presents a challenge for broader applicability. Future efforts aim to incorporate more geographically diverse datasets to address these limitations.

Dr. Wachinger and his team are advancing their work by obtaining large-scale CT datasets from the United States, which may help create a more representative sample and address the current lack of healthy reference cohorts. This study reinforces AI's role as a foundational infrastructure in quantitative radiology, transitioning beyond mere research outputs to practical tools for clinical application.