Retrospective T2 quantification from conventional weighted MRI of the prostate based on deep learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Accurate T2 quantification is crucial for prostate cancer diagnosis and characterization.
• Conventional methods for T2 mapping require specialized MRI sequences, limiting their routine clinical use.

Purpose of the Study:

• To develop and validate a deep learning model for retrospective T2 quantification using conventional T1- and T2-weighted MRI.
• To assess the feasibility of this deep learning approach for clinical prostate cancer assessment.

Main Methods:

• A U-Net based neural network was trained on multi-echo spin-echo data to predict T2 maps from conventional images.
• The model's performance was evaluated using metrics like SSIM, PSNR, and Pearson correlation.
• Retrospective T2 quantification was performed on high-risk prostate cancer and active surveillance cohorts.

Main Results:

• The deep learning model accurately generated T2 maps comparable to reference data, preserving tissue structures and contrast.
• Significant differences in T2 values were observed between tumor and non-tumor regions in prostate cancer patients (P < 0.001).
• Active surveillance patients who progressed showed lower tumor T2 values over time, distinguishing them from non-progressors (P = 0.010).

Conclusions:

• A deep learning method enables retrospective T2 mapping from conventional MRI, offering a non-invasive tool for prostate cancer evaluation.
• This approach has the potential to enhance prostate cancer diagnosis and characterization without necessitating extra MRI scans.