Selecting proper combination of mpMRI sequences for prostate cancer classification using multi-input convolutional...

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• Accurate classification of prostate cancer (PCa) is crucial for patient management.
• Multiparametric magnetic resonance imaging (mpMRI) is a key tool for PCa detection.
• Distinguishing clinically significant (CS) from clinically insignificant (CiS) PCa remains a challenge.

Purpose of the Study:

• To develop a deep learning model for automated classification of CS and CiS PCa using mpMRI.
• To identify optimal mpMRI sequences for PCa classification in different prostate zones.
• To enhance diagnostic accuracy and efficiency in PCa detection.

Main Methods:

• A multi-input selection network (MISN) was designed, processing nine mpMRI sequences.
• A pruning strategy created zone-specific networks (PZN for peripheral zone, TZN for transition zone).
• A penalized cross-entropy loss function was employed to balance sensitivity and specificity.

Main Results:

• The MISN achieved an Area Under the Curve (AUC) of 0.95 on the PROSTATEx dataset, outperforming existing methods.
• Zone-specific networks (PZN and TZN) demonstrated superior performance compared to the general MISN.
• The approach ranked first among over 1500 submissions in the PROSTATEx challenge.

Conclusions:

• Deep learning models can effectively classify prostate cancer using mpMRI.
• Strategic selection of mpMRI sequences enhances classification performance.
• This method offers a promising approach for improving PCa diagnosis.