Development of PDAC diagnosis and prognosis evaluation models based on machine learning

Area of Science:

• Oncology
• Biomedical Informatics
• Machine Learning

Background:

• Pancreatic ductal adenocarcinoma (PDAC) presents diagnostic challenges due to its aggressive nature and lack of early detection methods.
• Current serum biomarkers like CA19-9 have limited efficacy for early PDAC diagnosis.
• Machine learning (ML) and deep learning (DL) offer promising avenues for improving PDAC detection and patient management.

Purpose of the Study:

• To develop ML-based models for differential diagnosis of PDAC.
• To establish DL models for accurate prognosis assessment in PDAC patients.
• To utilize model predictions for personalized treatment recommendations and improved survival rates.

Main Methods:

• Utilized serum biomarker data and prognosis information from 117 PDAC patients.
• Employed ML models (Random Forest, Neural Network, SVM, GBM) for differential diagnosis, evaluated using accuracy, Kappa, ROC, sensitivity, and specificity.
• Applied COX regression and DeepSurv DL model for survival risk prediction, comparing performance via C-index and Log-rank test.

Main Results:

• ML models demonstrated effective PDAC diagnosis, with accuracies ranging from 76.97% to 84.21%.
• The DeepSurv DL model outperformed the COX model in survival risk prediction (C-indexes 0.738 training, 0.724 validation).
• Personalized treatment recommendations based on DeepSurv predictions showed potential for patient survival benefits.

Conclusions:

• Developed efficient ML and DL models for PDAC diagnosis and prognosis.
• The DeepSurv model proved superior for prognosis prediction, guiding personalized treatment strategies.
• The study supports the integration of ML/DL models in clinical management for enhanced PDAC patient outcomes.