Development of a Radiomic-Based Model Predicting Lymph Node Involvement in Prostate Cancer Patients

Area of Science:

• Radiology
• Oncology
• Medical Imaging Analysis

Background:

• Lymph node involvement (LNI) in prostate cancer (PCa) significantly impacts treatment decisions.
• While visual MRI interpretation aids LNI risk modeling, quantitative analysis offers potential for improved prediction.
• Current models for LNI risk prediction in PCa have limitations in accuracy.

Purpose of the Study:

• To develop and validate a novel LNI risk prediction model for prostate cancer.
• To leverage quantitative radiomic features from preoperative multimodal MRI for enhanced prediction.
• To compare the performance of the novel model against existing state-of-the-art LNI risk prediction models.

Main Methods:

• Retrospective analysis of 280 patients undergoing preoperative MRI and radical prostatectomy with lymph node dissection.
• Extraction of radiomic features from index tumor volumes on ADC corrected maps and T2 sequences.
• Development of a prediction model using a neural network (Multilayer Perceptron) combining clinical and radiomic features, with ComBat harmonization for inter-site heterogeneity.

Main Results:

• The proposed combined model achieved a C-Index of 0.89 in the testing set, outperforming established models (C-Indices ranging from 0.55 to 0.73).
• Radiomic features from preoperative MRI, when combined with clinical data via a neural network, demonstrated superior predictive performance.
• The study included 280 patients with a median age of 65.2 years and 18.2% LNI incidence.

Conclusions:

• Quantitative radiomic analysis of preoperative MRI, integrated with clinical data, significantly enhances lymph node involvement risk prediction in prostate cancer.
• The developed neural network-based model offers improved accuracy over current state-of-the-art models for LNI risk assessment.
• This approach holds promise for more precise treatment planning in prostate cancer management.