CT-based machine learning radiomics predicts Ki-67 expression level and its relationship with overall survival in resectable pancreatic ductal adenocarcinoma

Abstract

Background: The prognostic prediction of pancreatic ductal adenocarcinoma (PDAC) remains challenging. This study aimed to develop a radiomics model to predict Ki-67 expression status in PDAC patients using radiomics features from dual-phase enhanced CT, and integrated clinical characteristics to create a radiomics-clinical nomogram for prognostic prediction.

Methods: In this retrospective study, data were collected from 124 PDAC patients treated surgically at a single center, from January 2017 to March 2023. Patients were categorized according to the Ki-67 expression rate. Radiomics features were extracted from arterial and portal venous phase CT images using 3D Slicer v5.0.3. A radiomics model was formulated and validated to predict the Ki-67 expression, and a nomogram combining clinical indicators and the radiomics model was developed to predict 1, 2 and 3 year overall survival (OS).

Results: The optimal Ki-67 expression rate cutoff was identified as 50%, with significant OS differences. The developed radiomics model showed good predictive ability with area under the curves of 0.806 and 0.801 in the training and validation groups, respectively. High radiomics score, elevated carbohydrate antigen 19-9 (CA19-9), and receipt of adjuvant chemotherapy were identified as independent prognostic factors for OS. The radiomics-clinical nomogram accurately predicted 1, 2 and 3 year OS in PDAC patients.

Conclusions: The radiomics-clinical nomogram provides a non-invasive and efficient method for predicting Ki-67 expression and overall survival in PDAC patients, which could potentially guide clinical decision-making and improve patient outcomes.

Keywords: Ki-67; Nomogram; Overall survival; Pancreatic ductal adenocarcinoma; Radiomics.