TL;DR
This study introduces an MRI-based radiomics approach to differentiate between two rare pancreatic tumors: hypovascular non-functional pancreatic neuroendocrine tumors (hypo-NF-pNETs) and solid pseudopapillary neoplasms (SPNs). By analyzing quantitative imaging features from contrast-enhanced MRI, the researchers developed a nomogram that shows promising accuracy in distinguishing these tumors, which can be challenging to differentiate using conventional methods.
What Problem Does This Paper Address?
Accurately distinguishing between different types of pancreatic tumors is crucial for effective treatment and patient management. Solid pseudopapillary neoplasms (SPNs) and non-functional pancreatic neuroendocrine tumors (NF-pNETs) are both rare pancreatic tumors, but they have distinct prognoses and treatment strategies. While NF-pNETs can present with varied imaging characteristics, including hypovascular patterns, they can sometimes mimic SPNs on standard imaging. This diagnostic challenge is exacerbated by the low incidence of these tumors, leading to a scarcity of research on their differentiation. The paper aims to address this by developing a quantitative, objective method using radiomics to improve the diagnostic accuracy in differentiating hypovascular NF-pNETs from SPNs.
Methods
The study employed a retrospective, single-center design. Patients were categorized into two groups: those with pathologically confirmed SPNs and those with hypovascular NF-pNETs. Contrast-enhanced MRI (CE-MRI) was performed, and radiomics features were extracted from T1-weighted images across different phases: arterial, portal, and delayed.
Key methodological steps included:
- Image Segmentation: Regions of interest (ROIs) were manually drawn on the arterial phase of CE-MRI images by experienced radiologists and transferred to other phases. Inter- and intra-observer agreement for segmentation was assessed using the Intraclass Correlation Coefficient (ICC), with features having ICC > 0.75 being retained.
- Feature Extraction: The PyRadiomics package was used to extract a comprehensive set of radiomics features, including first-order, 2D shape, texture, and high-order features. Image preprocessing steps like voxel resampling, gray-level discretization, and intensity normalization were applied.
- Feature Selection: Two methods, max-relevance and min-redundancy (mRMR) and Least Absolute Shrinkage and Selection Operator (LASSO) regression, were used to select the most predictive features and reduce redundancy.
- Model Construction: Radiomics signatures were built by summing the selected features weighted by their coefficients. A clinic-radiomics nomogram was then constructed by integrating the best-performing radiomics signature (from the arterial phase) with significant clinical predictors identified through logistic regression analysis (age was found to be significant).
- Validation: The models were validated using a 7:3 training-validation split. Receiver Operating Characteristic (ROC) analysis was used to evaluate the diagnostic performance (AUC, sensitivity, specificity). Calibration curves and decision curve analysis (DCA) were employed to assess model calibration and clinical utility, respectively.
Datasets
The study utilized a dataset of 57 patients with SPNs and 22 patients with hypovascular NF-pNETs, totaling 79 patients. These patients were recruited from Changhai Hospital between October 2012 and September 2018. The dataset was split into a training cohort (70% of patients) and a validation cohort (30% of patients). No new benchmark datasets were introduced; the study relied on existing patient data.
Key Insights and Ideas
- Radiomics as an Objective Differentiator: The study highlights the potential of radiomics to provide objective, quantitative features from MRI that can surpass the subjectivity of traditional imaging interpretation for differentiating challenging pancreatic tumors.
- Arterial Phase Dominance: The arterial phase of contrast-enhanced MRI yielded the most robust radiomics signature for differentiating SPNs from hypo-NF-pNETs, suggesting that this phase captures critical tumor characteristics relevant to this distinction.
- Age as a Key Clinical Predictor: While radiomics features were powerful, the inclusion of patient age as a clinical predictor in the nomogram significantly enhanced its discriminatory power, underscoring the interplay between imaging biomarkers and demographic factors.
- Nomogram for Clinical Decision Support: The developed clinic-radiomics nomogram offers a practical tool that integrates both imaging-derived features and clinical information, aiming to assist clinicians in making more informed diagnostic decisions.
Key Results
- Radiomics Signatures Performance: Radiomics signatures showed strong discriminatory abilities. In the training cohort, AUCs ranged from 0.871 (precontrast T1WI) to 0.996 (delayed phase). In the validation cohort, AUCs were 0.853 (precontrast T1WI) to 0.907 (arterial phase).
- Arterial Phase Signature: The radiomics signature from the arterial phase achieved AUCs of 0.978 (training) and 0.907 (validation), demonstrating its high diagnostic value.
- Clinic-Radiomics Nomogram Performance: The nomogram, combining age and the arterial phase radiomics signature, achieved AUCs of 0.965 (training) and 0.920 (validation).
- Clinical Utility: Decision curve analysis indicated that the nomogram provided a higher net benefit compared to treat-all or treat-none strategies.
- Demographic Differences: SPNs were found to occur in younger patients (mean age 34.2 years) compared to hypo-NF-pNETs (mean age 47.2 years), and showed a female preponderance (77.2% vs. 50%).
Limitations and Future Work
The authors acknowledge several limitations:
- Retrospective Design and Single Institution: This retrospective, single-institution study may be subject to selection bias.
- Small Sample Size: The relatively small number of patients, particularly for hypo-NF-pNETs, is attributed to the low incidence of these tumors and strict inclusion criteria.
- Nomogram Superiority: The clinic-radiomics nomogram did not show statistically significant superiority over the arterial phase radiomics signature alone.
Future work suggested includes:
- Multicenter Prospective Studies: Conducting multicenter studies with larger patient cohorts is recommended to validate these findings and confirm the potential value of MR radiomics.
- Exploration of Other Clinical Factors: Further investigation into other clinical factors that could improve the nomogram's performance is warranted.
- Prospective Validation: The nomogram needs to be modified and prospectively validated to assess its real-world clinical utility.
Summary updated