Non-Invasive PNETs Grading Using CT Radiomics and Machine Learning

Abstract

Abstract Purpose The purpose is to determine the most effective machine learning method for identifying pathological grades of pancreatic neuroendocrine tumors (PNETs). This will be achieved by analyzing contrast-enhanced computed tomography scans of both arterial and portal phases. This investigation aims to provide clinicians with an efficient and reliable tool for accurately identifying PNETs pathological grades. Materials and Methods An investigation was conducted on a cohort of 100 patients diagnosed with pancreatic neuroendocrine tumors. Definitive pathological diagnoses and grading results were obtained for all patients. Pyradiomics, an open-source Python package, extracted radiomics features from CT images obtained in arterial and portal phases. These features were subsequently utilized in different machine learning classifiers. The classification model's performance was assessed using sensitivity, specificity, area under the curve (AUC), and accuracy metrics. Result Our analysis demonstrates that combining CT-based radiomics features with a multi-algorithm machine learning approach can accurately identify the pathological grades of pancreatic neuroendocrine tumors. The most effective predictive performance in the arterial phase was observed with the combination of Arterial_RFE and LR, with an AUC of 0.68. Meanwhile, in the portal phase, the combination of Portal_RFE and KNN demonstrated the highest predictive performance with an AUC of 0.76. Conclusion The application of CT radiomics features, augmented with machine learning, has shown promising results in determining the pathological grade of pancreatic neuroendocrine tumors. This approach can further contribute to the classification of PNET patients into grade 1 and grade 2/3 categories based on arterial and portal phases, enabling clinical decision-making.