Radiomics-Based Machine Learning Model for Differential Diagnosis of Pancreatic Cystic Neoplasms

Introduction: Accurate differential diagnosis of pancreatic cystic lesions is crucial for selecting the optimal management strategy and for timely identification of lesions with malignant potential. A key objective of radiology is the development and implementation of quantitative imaging approaches that can improve diagnostic performance and reduce reliance on subjective image interpretation. Purpose: To evaluate the value of magnetic resonance imaging (MRI) and MRI-based radiomic analysis for the differential diagnosis of pancreatic cystic neoplasms, and to develop machine-learning radiomic models for predicting the malignant potential of pancreatic cystic lesions. Materials and methods: In this retrospective study, MRI examinations of 67 patients with surgically and pathologically confirmed pancreatic cystic neoplasms were analyzed. Lesion segmentation and radiomic feature extraction were performed by a radiologist experienced in abdominal imaging. Seven supervised machine-learning models were trained. Model performance was assessed using area under the receiver operating characteristic curve (ROC-AUC), precision–recall AUC (PR-AUC), accuracy, sensitivity, specificity, and F1-score. The best-performing model was selected based on ROC-AUC. Model interpretability was evaluated using SHapley Additive exPlanations (SHAP). Results: The Random Forest classifier showed the best performance (ROC-AUC = 0.83), followed by LightGBM (ROC-AUC = 0.77). SHAP analysis highlighted the radiomic features with the greatest impact on the model outputs. Conclusion: MRI-based radiomics may support risk stratification of pancreatic cystic lesions. Further studies with larger cohorts are needed to confirm these findings and improve model generalizability.

1. Skulskiy SK, Ratnikov VA, Lubashev YA, et al. MRI diagnosis pancreas cystics neoplasms at the stages of a medical examination. Vrach. 2022;33(11):40-47. (In Russ.) https://doi.org/10.29296/25877305-2022-11-07

2. Ardeshna DR, Cao T, Rodgers B, et al. Recent advances in the diagnostic evaluation of pancreatic cystic lesions. World J Gastroenterol. 2022;28(6):624-34. https://doi.org/10.3748/wjg.v28.i6.624

3. Rogowska J, Semeradt J, Durko Ł, Małecka-Wojciesko E. Diagnostics and Management of Pancreatic Cystic Lesions-New Techniques and Guidelines. J Clin Med. 2024 Aug 8;13(16):4644. https://doi.org/10.3390/jcm13164644

4. Kondratyev EV, Shmeleva SA, Ustalov AA, et al. Theoretical basics of abdominal CT radiomics: a review. Diagnostic Radiology and Radiotherapy. 2025;16(1):33-46. (In Russ.). https://doi.org/10.22328/2079-5343-2025-16-1-33-46

5. Ștefan PA, Lupean RA, Lebovici A, et al. Quantitative MRI of Pancreatic Cystic Lesions: A New Diagnostic Approach. Healthcare (Basel). 2022;10(6):1039. https://doi.org/10.3390/healthcare10061039

6. Yao L, Zhang Z, Demir U, et al. Radiomics Boosts Deep Learning Model for IPMN Classification. Mach Learn Med Imaging. 2023;14349:134 143. https://doi.org/10.1007/978-3-031-45676-3_14

7. Zwanenburg A, Vallières M, Abdalah MA, et al. The Image Biomarker Standardization Initiative: Standardized Quantitative Radiomics for High-Throughput Image-based Phenotyping. Radiology. 2020;295(2):328-338. https://doi.org/10.1148/radiol.2020191145

8. Lambin P, Woodruff HC, Mali SA, et al. Radiomics Quality Score 2.0: towards radiomics readiness levels and clinical translation for personalized medicine. Nat Rev Clin Oncol. 2025;22(11):831-846. https://doi.org/10.1038/s41571-025-01067-1

9.

10.