JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

Comparison of multiple machine learning models for predicting prognosis of pancreatic ductal adenocarcinoma based on contrast-enhanced CT radiomics and clinical features

  • 0Department of Hepatopancreatobiliary Surgery, The First Affiliated Hospital, Fujian Medical University, Fuzhou, Fujian, China.
Frontiers in Oncology +

|

November 28, 2024

|

November 28, 2024

View abstract on PubMed

Summary

This summary is machine-generated.

This study combined clinical features and radiomics using machine learning to predict pancreatic cancer prognosis. The developed model shows excellent accuracy, offering a noninvasive tool for clinical decision-making.

Area Of Science

  • Oncology
  • Radiology
  • Data Science

Background

  • Pancreatic ductal adenocarcinoma (PDAC) poses significant challenges in prognosis prediction.
  • Accurate prognostic models are crucial for effective clinical decision-making in PDAC management.

Purpose Of The Study

  • To evaluate the prognostic potential of combining clinical features and radiomics with machine learning (ML) algorithms in PDAC.
  • To develop and validate a predictive model for PDAC prognosis.

Main Methods

  • 116 PDAC patients were divided into training and validation cohorts.
  • Seven ML algorithms were integrated into 43 combinations to construct radiomics models using arterial phase (AP), venous phase (VP), and combined (AP+VP) images.
  • Cox regression analyses were used to identify prognostic indicators and build a combined model integrating radiomics scores (Radscore) and clinical features.

Main Results

  • The Lasso+StepCox algorithm with AP+VP radiomics features yielded the best radiomics model (C-indices: 0.742 training, 0.722 validation).
  • A combined model incorporating sex, TNM stage, systemic inflammation response index, and AP+VP-Radscore achieved higher C-indices (0.764 training, 0.746 validation).
  • The combined model demonstrated good consistency and net benefit in calibration curves and decision curve analysis.

Conclusions

  • A combined model integrating clinical features and AP+VP-Radscore, selected via ML, shows excellent prognostic prediction ability for PDAC.
  • This approach offers a potentially noninvasive and effective method for clinical decision-making in PDAC.

Keywords: