Prostate cancer and benign prostatic hyperplasia lesions segmentation using diffusion kurtosis imaging, T2*, and R2*...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Prostate cancer diagnosis relies heavily on multiparametric MRI (mpMRI).
• Accurate segmentation of prostate lesions is crucial for treatment planning and monitoring.
• Existing segmentation methods may struggle with diverse MRI-derived parametric maps.

Purpose of the Study:

• To develop and evaluate a deep learning segmentation framework for prostate lesions across multiple MRI acquisitions and derived parametric maps.
• To compare the performance of the segmentation model across different MRI-derived images.
• To identify MRI-derived images that provide the most discriminative information for accurate prostate lesion identification.

Main Methods:

• A U-Net++ deep learning model was trained on mpMRI data from 51 patients.
• Images included T2-weighted imaging (T2WI), apparent diffusion coefficient (ADC) map, diffusion kurtosis imaging (DKI)-derived maps (D, K), and T2*-weighted imaging-derived maps (T2*, R2*).
• Segmentation performance was assessed using Dice similarity coefficient and Intersection over Union (IoU), with manual annotations by expert radiologists.

Main Results:

• The deep learning model achieved high segmentation performance across all tested MRI-derived images.
• The K map (IoU: 0.9504, Dice: 0.9744) and T2WI (IoU: 0.9250, Dice: 0.9604) demonstrated the highest segmentation accuracy.
• IoU values ranged from 0.8559 (D map) to 0.9504 (K map), and Dice scores ranged from 0.9211 (D map) to 0.9744 (K map).

Conclusions:

• The proposed deep learning framework effectively delineates prostate lesions on various mpMRI-derived images.
• K map and T2WI offer the most discriminative information for accurate prostate lesion segmentation.
• Combining advanced deep learning with optimized MRI-derived images enhances diagnostic precision for prostate cancer.