MPCaD: a multi-scale radiomics-driven framework for automated prostate cancer localization and detection

Area of Science:

• Radiology
• Medical Imaging
• Oncology

Background:

• Quantitative radiomic features from multi-parametric magnetic resonance imaging (MP-MRI) offer valuable data for prostate cancer detection and localization.
• Current algorithms often use simplified models, but the complex nature of prostate tumors in MP-MRI necessitates advanced frameworks.

Purpose of the Study:

• To present MPCaD, a novel Multi-scale radiomics-driven framework for Prostate Cancer Detection and localization.
• To integrate multi-scale radiomic feature models and a priori knowledge for improved accuracy.

Main Methods:

• MPCaD employs a statistical texture distinctiveness strategy for initial tumor candidate localization using voxel-resolution features.
• Tumor region classification is performed using a region-resolution feature model.
• A conditional random field framework refines localization using voxel-resolution relative ADC features.

Main Results:

• The framework was evaluated on clinical prostate MP-MRI data from 30 patients.
• MPCaD demonstrated enhanced separability between cancerous and healthy prostate tissue.
• The proposed framework outperformed individual quantitative radiomics models in prostate cancer detection.

Conclusions:

• Quantitative radiomic features from MP-MRI are effective for prostate cancer detection and localization.
• The novel MPCaD framework offers a more sophisticated approach to leveraging MP-MRI data for improved prostate cancer diagnosis.