Habitat Imaging Model Enhances Noninvasive Risk Stratification of Subsolid Lung Nodules

A new multicenter study published in the American Journal of Roentgenology (AJR) demonstrates that habitat imaging, a method that quantifies spatial heterogeneity within lesions, offers a powerful and interpretable tool for stratifying the risk of subsolid nodules (SSNs) detected during lung cancer screening.

Led by Jieke Liu, MD, of the Sichuan Clinical Research Center for Cancer and the University of Electronic Science and Technology in Chengdu, China, the research team found that a ternary-classification habitat model performed significantly better than a conventional 2D model for predicting the invasiveness and grade of lung adenocarcinomapresenting as SSNs on low-dose chest CT (LDCT).

“The ternary-classification habitat model for invasiveness and grade of lung adenocarcinoma presenting as a subsolid nodule on low-dose chest CT performed significantly better than the 2D model,” Liu wrote. “Its performance was not significantly different from radiomic and combined models.”

The study evaluated 747 patients (median age 56 years; 241 male, 506 female) with 834 surgically resected lung adenocarcinomas identified as SSNs on LDCT between July 2018 and May 2023. Data from one center were divided into training (n=440) and internal testing (n=189) sets, while adenocarcinomas from three additional centers formed an external validation cohort (n=205).

Researchers classified tumors into three categories: noninvasive adenocarcinoma, grade 1 invasive adenocarcinoma (IAC), and grade 2 or 3 IAC. Using multivariable multinomial logistic regression, they developed four classification models:

• 2D model: based on diameter and consolidation-to-tumor ratio (CTR).
• Habitat model: based on volume and the proportion of attenuation-defined subregions.
• Radiomic model: based on extracted radiomic features.
• Combined model: incorporating both habitat and radiomic features.

In the external test set, the models achieved the following macro-aggregated AUCs for ternary classification:

• 2D model: 0.871
• Habitat model: 0.919
• Radiomic model: 0.924
• Combined model: 0.926

One example from the study involved a 53-year-old woman with a pure ground-glass nodule, later diagnosed as minimally invasive adenocarcinoma. Habitat imaging segmented the lesion into distinct attenuation-based subregions, with calculated volumes and volume ratios illustrating the model’s capacity to capture internal heterogeneity.

The habitat model’s major advantage lies in its interpretability—its reliance on volume-based features that are intuitive and clinically meaningful, unlike some radiomic features that can function as “black box” metrics. By visualizing and quantifying the spatial heterogeneity of SSNs, the model provides radiologists and oncologists with a clearer framework for noninvasive risk stratification.

“The habitat model’s combination of interpretability and diagnostic performance support its utility for noninvasive risk stratification of SSNs encountered during lung cancer screening,” the study authors concluded.

As lung cancer screening with LDCT becomes more widespread, tools like habitat imaging may help refine patient management strategies, guiding decisions about surveillance versus intervention, and reducing unnecessary procedures while ensuring early detection of aggressive disease.