Computer-aided classification of prostate cancer grade groups from MRI images using texture features and stacked...

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• Accurate grading of prostate cancer (PCa) is crucial for treatment decisions.
• Multi-parametric magnetic resonance imaging (mpMRI) offers valuable biomarkers for PCa assessment.
• Current methods for determining PCa Grade Group (GG) can be improved with advanced computational techniques.

Purpose of the Study:

• To develop and evaluate a novel method for classifying prostate cancer into 5 Grade Groups (GG) using mpMRI biomarkers.
• To leverage deep learning, specifically stacked sparse autoencoders (SSAE), for feature extraction from mpMRI data.
• To assess the performance of the proposed method in a competitive challenge setting.

Main Methods:

• Utilized hand-crafted texture features from T2 Weighted (T2W), Apparent Diffusion Coefficient (ADC), and high B-Value Diffusion-Weighted (BVAL) MRI images.
• Employed a deep network of stacked sparse autoencoders (SSAE) for high-level feature extraction.
• Classified extracted features using a softmax classifier (SMC) on the PROSTATEx-2 2017 challenge dataset.

Main Results:

• The method achieved first place in the PROSTATEx-2 2017 challenge, outperforming 43 other methods.
• Achieved a quadratic-weighted Kappa score of 0.2772 on the test dataset.
• Demonstrated a Positive Predictive Value (PPV) of 80% for predicting PCa with GG > 1, even with an imbalanced training dataset.

Conclusions:

• The developed SSAE-based method shows significant potential for accurate prostate cancer Grade Group classification from mpMRI.
• This pioneering approach provides a strong foundation for future research and refinement in AI-driven PCa diagnostics.
• The method's success in a competitive challenge highlights its efficacy and robustness in classifying PCa into 5 grade groups.