AI-based CT-derived body composition quantification, particularly baseline SMVI and dynamic changes in SMVI and SAVI during neoadjuvant immunochemotherapy, are independently associated with pathological complete response in NSCLC, and incorporating these biomarkers into predictive models improves performance beyond clinical variables alone.
Key Findings
Results
High baseline skeletal muscle volume index (SMVI) was independently associated with pathological complete response (pCR) to neoadjuvant immunochemotherapy in NSCLC.
Multivariable analysis yielded an odds ratio of 2.22 for high baseline SMVI and pCR.
The study included 657 patients with a mean age of 61.3 years, 87.4% men.
pCR rates were 39.7% (training), 38.4% (internal validation), and 34.9% (external validation cohorts).
The study was a multicenter retrospective study conducted in China between July 2019 and July 2024.
Results
Each 1% relative increase in SMVI during neoadjuvant immunochemotherapy was associated with a 16% higher likelihood of pCR.
The odds ratio for each 1% relative increase in SMVI (%ΔSMVI) was OR = 1.16.
This association was identified in multivariable analysis.
SMVI change was measured between pre- and post-treatment CT scans using automated volumetric body composition segmentation.
Results
Every 10% relative increase in subcutaneous adipose volume index (SAVI) during neoadjuvant immunochemotherapy improved pCR probability.
The odds ratio for each 10% relative increase in SAVI (%ΔSAVI) was OR = 1.56.
This association was identified in multivariable analysis.
Body composition metrics included skeletal muscle, intermuscular, visceral, and subcutaneous adipose volume index.
Results
A machine learning model integrating clinical variables, baseline SMVI, %ΔSMVI, and %ΔSAVI demonstrated significantly better discrimination than models using clinical variables alone.
The improvement in discrimination was statistically significant (p < 0.05) in all cohorts.
In the internal validation cohort, the model achieved a sensitivity of 62.1% and specificity of 66.7%.
In the external validation cohort, the model achieved a sensitivity of 52.8% and specificity of 74.7%.
The model incorporated both baseline and dynamic (percentage change) body composition metrics.
Methods
An automated AI-based pipeline was used for three-dimensional CT-derived body composition segmentation spanning T1-T12 vertebral levels.
Pre- and post-treatment CT scans were used for automated T1-T12 localization and volumetric body composition segmentation.
Metrics extracted included skeletal muscle, intermuscular, visceral, and subcutaneous adipose volume index (SAVI).
Both absolute baseline values and percentage changes between scans were computed.
The approach was applied across multiple centers in a retrospective design.
Background
Tumor-intrinsic biomarkers alone were considered insufficient to predict pCR to neoadjuvant immunochemotherapy in NSCLC, motivating evaluation of body composition as a complementary biomarker.
The study rationale was that tumor-intrinsic biomarkers alone 'insufficiently predict pathological complete response (pCR) to neoadjuvant immunochemotherapy (NICT) in non-small cell lung cancer.'
Body composition metrics were described as 'modifiable biomarkers,' suggesting potential for intervention.
The study evaluated AI-based body composition as providing 'complementary predictive value.'
Huang Y, Wei Z, Ye G, Cui Y, Li C, Wu G, et al.. (2026). Automated CT-derived body composition predicts pathologic response to neoadjuvant immunotherapy in non-small cell lung cancer.. Cancer letters. https://doi.org/10.1016/j.canlet.2025.218229