Long-Term Prognosis of Resected Stage 1 Lung Adenocarcinoma With Radiomics, Regionality, and Histopathologic...

The discovery of adenocarcinoma (AC) spectrum nodules often leads to a pathway of ongoing radiologic observation, tissue biopsy, ablative therapies, or surgical excision. This complex decision-making process may be aided by machine learning. The Score Indicative of Lung Cancer Aggression was developed and validated as an evolution of previously developed radiomics models to predict the degree of histologic tumor invasion in AC and to predict patient survival. We now investigate if radiomics classification can help predict long-term outcomes after stage I AC resection. Secondarily, we explore if the geographic location of stage 1 AC nodules correlates with local invasion or long-term outcomes.

Can radiomics predict long-term outcomes in stage 1 lung adenocarcinoma after curative surgical resection?

A total of 196 patients with early stage AC who underwent curative surgical resection were studied over a follow-up median period of 12 years. Kaplan-Meier analysis of disease-free survival was performed on patients whose index nodule was categorized by radiomics. Local invasion of tumor was analyzed using radiomics and histopathologic assessment. The correlation of these metrics relating to the nodule's geographic location was explored. Finally, disease-free survival analysis based on the location of the index nodule was performed.

Patients whose stage I AC nodules were classified as poor before resection by our radiomics tool had worse long-term prognosis (> 5 years) than those classified as intermediate or good. AC nodules in lower lung regions were proportionately more invasive than nodules in the upper lung regions, both based on radiomics and histopathologic review. However, that did not translate to a disease-free survival advantage for upper lobe nodules that were resected.

Our results indicate that radiomics helps prognosticate long-term outcomes in patients with resected stage I AC. Furthermore, histopathologic invasion detected by radiomics seems to vary depending on geographic location of the nodule in the lung.