A self-tuned graph-based framework for localization and grading prostate cancer lesions: An initial evaluation based...

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Multiparametric magnetic resonance imaging (mpMRI) is crucial for prostate cancer detection.
• The Prostate Imaging-Reporting and Data System (PI-RADS) standardizes mpMRI reporting.
• Manual PI-RADS assessment is subjective and time-consuming.

Purpose of the Study:

• To develop and validate a self-tuned graph-based model for automated PI-RADS rating prediction.
• To improve the objectivity and efficiency of prostate cancer lesion assessment from mpMRI.

Main Methods:

• A self-tuned graph-based model was developed using 34 pixel-level features from T2W, ADC, and DCE MRI sequences.
• The model automatically tuned edge and feature weights for PI-RADS score prediction.
• Lesion localization performance was evaluated on mpMRI datasets from 12 patients.

Main Results:

• The algorithm achieved sensitivity, specificity, and accuracy ranging from 65-77%, 86-93%, and 85-88%, respectively.
• Performance was comparable to previous studies using non-clinical T2 maps.
• The model processed an axial image for PI-RADS score distribution in just 10 seconds.

Conclusions:

• The developed self-tuned graph-based model demonstrates efficient and accurate PI-RADS rating prediction.
• This automated approach offers a promising tool for clinical prostate MR analysis.
• Further validation with larger patient cohorts is recommended for clinical implementation.