🤖 Generated with AI assistance · Does not replace individualized medical advice.
AI Act Transparency Notice
This content was generated with the assistance of artificial intelligence tools. It is intended for informational and educational purposes only and does not constitute medical advice, diagnosis, or treatment. Always seek the advice of a qualified health provider with any questions regarding a medical condition.
Topologically Distinct Intratumoral Heterogeneity Scores Predict High-Risk Lung Adenocarcinoma Subtypes
Key Facts
| Metric | Value |
|---|---|
| Study Design | Multicenter retrospective cohort (4 institutions) + External Validation (TCIA) |
| Sample Size | 1,051 patients with invasive lung adenocarcinoma (IAC) |
| Primary Endpoint | Preoperative identification of high-risk pathological grades (micropapillary/solid) |
| Best Model Performance | Stacking Ensemble Classifier: AUC 0.875 PMID: 42709716 |
Why It Matters / Context
Invasive lung adenocarcinoma (IAC) is a heterogeneous disease where histological subtypes significantly drive prognosis. Specifically, micropapillary and solid-predominant patterns are associated with higher rates of lymph node metastasis, recurrence, and poor overall survival compared to lepidic or acinar patterns PMID: 42709716. Currently, definitive histological classification requires surgical resection and pathological examination, meaning patients are often subjected to invasive procedures before their specific risk profile is fully understood.
Preoperative risk stratification is critical for surgical planning, particularly when deciding between lobectomy and sublobar resection (segmentectomy or wedge resection). If a patient has a high probability of harboring a high-risk subtype, a more extensive resection may be warranted to ensure adequate margins and lymph node sampling. Conversely, low-risk patients might be candidates for less invasive approaches.
Traditional radiomics features often capture texture and shape but may fail to account for the spatial distribution of these features within the tumor volume. This study introduces “topologically distinct” ITH scores, which integrate local radiomics with global pixel distribution patterns. By extending this from 2D (largest slice) to 3D (entire tumor volume), the model captures the volumetric complexity of tumor heterogeneity, which is a known driver of aggressive tumor biology and treatment resistance PMID: 42709716.
Details
The study developed two complementary ITH scores: a 2D score based on the largest cross-sectional CT slice and a 3D score extending quantification across the entire tumor volume. These scores were combined with clinicoradiological features to train six base machine learning classifiers, which were then integrated into a final stacking ensemble classifier.
The following table summarizes the comparative performance of the different modeling approaches:
| Model Type | Input Features | AUC (Area Under ROC Curve) |
|---|---|---|
| Stacking Ensemble (Final) | Clinicoradiological + 2D ITH + 3D ITH | 0.875 PMID: 42709716 |
| Radiomics-Only Model | Radiomics features | 0.834 PMID: 42709716 |
| Clinicoradiological-Only Model | Clinical + Radiological features | 0.792 PMID: 42709716 |
Interpretability and Validation:
SHapley Additive exPlanations (SHAP) analysis was used to evaluate model interpretability. The 3D ITH score was identified as the most influential contributor to the model’s output, suggesting that volumetric heterogeneity is a stronger predictor of high-risk histology than 2D slice-based metrics or standard clinical variables PMID: 42709716.
External validation was performed using an independent dataset from The Cancer Imaging Archive (TCIA). This validation confirmed that higher 3D ITH scores were associated with:
1. More aggressive tumor biology.
2. Specific genomic features linked to poor outcomes.
3. Shorter recurrence-free survival (RFS).
4. Shorter overall survival (OS) PMID: 42709716.
Limitations
⚠️ Critical Limitations* Retrospective Design: The primary cohort is retrospective, introducing potential selection bias and confounding factors that cannot be fully controlled for.
* Single Imaging Modality: The model relies exclusively on CT imaging. It does not incorporate PET-CT metabolic data or MRI, which may provide complementary information on tumor biology.
* Generalizability of “High-Risk” Definition: The model predicts micropapillary or solid-predominant patterns. It does not distinguish between these two subtypes or predict other adverse features like vascular invasion or pleural invasion directly.
* Scanner Variability: While multicenter, the study does not explicitly detail the harmonization of CT acquisition parameters (kVp, mAs, slice thickness) across the four institutions, which can impact radiomic feature stability.
* Lack of Prospective Validation: The model has not yet been tested in a prospective clinical trial to confirm its real-world utility in guiding surgical decision-making.
Clinical Takeaway
Clinicians should consider that 3D intratumoral heterogeneity scores derived from standard preoperative CT scans may serve as a non-invasive biomarker to identify patients with invasive lung adenocarcinoma who are at high risk for aggressive histological subtypes, potentially informing the extent of surgical resection.
Full Reference
Lin S, Mao Y, He Q, Qi W, Zhang S, Li W. Topologically distinct intratumoral heterogeneity scores for predicting high-risk pathological grades in invasive lung adenocarcinoma: A multicenter study across four institutions. PLOS Digital Health. 2026. PMID: 42709716 [DOI: 10.1371/journal.pdig.0001646]