Effect of Accurate Segmentation of Prostate Areas on Radiomics based model for Prostate Cancer Classification.

Area of Science:

• Medical Imaging
• Oncology
• Radiology

Background:

• Prostate cancer (PCa) is a leading cancer in men, necessitating non-invasive diagnostic tools.
• Multiparametric MRI (mpMRI) combined with radiomics offers potential for PCa detection and stratification.
• Current radiomics models often focus on lesion evaluation, which is time-consuming and challenging.

Purpose of the Study:

• To investigate the impact of radiomics from different prostate gland regions on classification performance for clinically significant prostate cancer (csPCa).
• To determine the optimal imaging modality and region for radiomic analysis in PCa diagnosis.

Main Methods:

• Analysis of T2-weighted and diffusion-weighted images (DWI) from 80 patients.
• Extraction of radiomics features from the whole gland, transition zone, periphery zone, and areas of predefined widths.
• Feature selection using recursive feature elimination and a voting strategy, followed by training and testing of two machine learning models.

Main Results:

• The highest diagnostic accuracy of 81.9% (±7.8) was achieved using radiomics from the transition and periphery zones.
• Diffusion-weighted imaging (DWI) with a high b-value (800) was identified as the most informative imaging modality.
• The study highlights the importance of specific anatomical regions within the prostate for radiomic analysis.

Conclusions:

• Radiomics analysis focused on specific prostate zones, particularly the transition and periphery, can significantly improve diagnostic accuracy for csPCa.
• High b-value DWI is crucial for informative radiomic feature extraction in PCa diagnosis.
• Development of accurate segmentation algorithms is essential for robust radiomics models in csPCa diagnosis and management.