🤖 Generated with AI assistance · Does not replace individualized medical advice.
AI Act Transparency Notice
This content was generated with the assistance of artificial intelligence to summarize and structure scientific data. It is intended for informational purposes only and does not constitute medical advice, diagnosis, or treatment. Always seek the advice of a qualified healthcare provider with any questions regarding a medical condition.
Ultrasound Habitat Radiomics Predicts Invasive Breast Cancer with DCIS Component
Key Facts
| Metric | Value |
|---|---|
| Study Design | Dual-center retrospective cohort |
| Total Patients | 1,063 |
| Best Model AUC (Validation) | 0.910 (95% CI: 0.883–0.936) |
| Primary Outcome | Preoperative prediction of IBC-DCIS |
Why It Matters / Context
Accurate preoperative differentiation between pure ductal carcinoma in situ (DCIS) and invasive breast cancer with a DCIS component (IBC-DCIS) is critical for surgical planning. IBC-DCIS often requires more extensive surgical margins and potentially different adjuvant therapy regimens compared to pure DCIS. While mammography and MRI are standard, ultrasound is widely accessible, cost-effective, and increasingly used for breast imaging. However, conventional ultrasound relies heavily on subjective visual assessment, which can lack specificity for distinguishing subtle invasive components within a predominantly in-situ lesion.
Radiomics, the extraction of high-dimensional quantitative features from medical images, offers a potential solution to quantify intratumoral heterogeneity. Traditional radiomics often treats the tumor as a homogeneous region of interest (ROI). However, breast tumors are biologically heterogeneous. The “habitat” approach partitions the tumor into distinct subregions based on texture and intensity, reflecting different microenvironments (e.g., necrotic, fibrotic, or highly cellular areas). This study advances the field by applying this subregional analysis to ultrasound data, aiming to capture the spatial complexity that may signal the presence of invasive foci.
Details
The study enrolled 1,063 pathologically confirmed breast cancer patients from two centers. The cohort was split into a training set (n = 637) and an external validation set (n = 426) PMID: 42701070. Tumor regions of interest were manually delineated on two-dimensional ultrasound images. To account for intratumoral heterogeneity, these ROIs were partitioned into three habitat subregions using unsupervised clustering. Radiomics features were extracted from both the whole tumor and each individual subregion.
Feature selection was rigorous, employing Pearson correlation analysis to remove redundant features and Least Absolute Shrinkage and Selection Operator (LASSO) regression to identify the most predictive variables. Multiple machine learning algorithms were constructed and compared. The Support Vector Machine (SVM) based combined model, which integrated features from the whole tumor and the three habitat subregions, demonstrated superior performance.
In the external validation cohort, the SVM combined model achieved an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.910 (95% CI: 0.883–0.936) PMID: 42701070. Statistical comparison using the DeLong test confirmed that this combined model significantly outperformed models based solely on imaging features or single-region radiomics (p < 0.05) PMID: 42701070.
To ensure robustness against potential class imbalance, the authors evaluated the model using a comprehensive suite of complementary metrics, including sensitivity, specificity, positive and negative predictive values, balanced accuracy, F1-score, Precision-Recall AUC (PR-AUC), and Brier score. The model showed acceptable calibration and provided clinical net benefit across a broad range of threshold probabilities, as evidenced by decision curve analysis PMID: 42701070.
| Model Type | Performance Metric | Result (External Validation) |
|---|---|---|
| SVM Combined (Habitat) | AUC | 0.910 (95% CI: 0.883–0.936) |
| Imaging-based | AUC | Significantly lower (p < 0.05) |
| Single-Region Radiomics | AUC | Significantly lower (p < 0.05) |
Limitations
Critical Appraisal:
* Retrospective Design: The study is retrospective, introducing potential selection bias and limiting causal inference.
* Manual Delineation: Tumor ROIs were manually delineated, which introduces inter-observer variability and reduces reproducibility compared to automated segmentation.
* 2D Ultrasound: The analysis relied on two-dimensional ultrasound images. Three-dimensional (3D) ultrasound or volumetric data might capture more comprehensive spatial information.
* External Validation Scope: While externally validated at a second center, the generalizability to other populations, ultrasound equipment manufacturers, or imaging protocols remains to be tested.
* Lack of Prospective Data: No prospective clinical trial data is available to confirm that using this model for surgical planning improves patient outcomes.
Clinical Takeaway
Clinicians should consider that ultrasound habitat radiomics offers a promising, non-invasive adjunct for preoperative risk stratification of IBC-DCIS, potentially aiding in surgical planning, though prospective validation is required before routine clinical implementation.
Full Reference
Fan L, Wu H, Wu Y, Wu S, Zhang C, Zhu X. Ultrasound Habitat Radiomics for Preoperative Prediction of Invasive Breast Cancer With a DCIS Component: A Dual-center Retrospective Study. Ultrasound in Medicine & Biology. 2026. PMID: 42701070 [DOI: 10.1016/j.ultrasmedbio.2026.06.006]