Multi-task Bayesian model combining FDG-PET/CT imaging and clinical data for interpretable high-grade prostate cancer...

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Prostate cancer (PCa) prognosis prediction is crucial for treatment planning.
• Accurate prediction of high-grade PCa (Gleason ≥ 8) outcomes remains challenging.
• Current methods often lack integration of multi-modal data and dynamic prediction capabilities.

Purpose of the Study:

• To develop and validate a fully automatic multi-task Bayesian model (BSN) for predicting high-grade PCa prognosis.
• To leverage pre-prostatectomy FDG-PET/CT images and clinical data for enhanced predictive accuracy.
• To introduce novel features like automated segmentation, uncertainty quantification, and dynamic predictions.

Main Methods:

• Development of the Bayesian Sequential Network (BSN), a multi-task Bayesian model.
• Integration of FDG-PET/CT imaging and clinical data for 295 patients.
• Implementation of classification and multiple survival prediction tasks (e.g., lymph node invasion, recurrence-free survival).

Main Results:

• BSN demonstrated superior performance compared to traditional nomograms across most prognostic tasks.
• The model successfully performed automated prostate segmentation and provided uncertainty quantification.
• Dynamic predictions, refining long-term prognosis using short-term outcomes, were introduced.

Conclusions:

• The BSN model shows significant promise for improving prognostic accuracy in high-grade prostate cancer.
• It effectively utilizes multi-modal data, including imaging, for personalized risk stratification.
• BSN can aid in identifying patients requiring treatment intensification, optimizing therapeutic strategies.