Background:

Prostate MRI is commonly performed on 3 Tesla (T) scanners due to improved signal-to-noise ratio (SNR) compared to 1.5T, as high SNR is important for prostate cancer detection and staging. However, 3T MRI is associated with increased artifact and higher costs. 1.5T MRI using machine learning-based reconstruction algorithm (DLR) has the potential to offer comparable image quality by reducing artifacts. Furthermore, 1.5T MRI with DLR could reduce scan times and costs associated with prostate MRI.

Methods:

A retrospective review found men who underwent 3T MRI without DLR and 1.5T MRI with DLR for prostate cancer evaluation between January 2024 and April 2024. 20 consecutive men with prostate lesions in each MRI system were included in this study. For each patient, axial T2W high resolution images with slice thickness of 3 mm were examined. A circular region-of-interest (ROI) was placed over the lesion (SI1) on the axial image demonstrating maximal lesion diameter and a second ROI of similar size was placed at the ipsilateral gluteus muscle (SI2) on the same image. Four ROIs were drawn at four corners of the image to calculate background noise; mean standard deviation from these 4 corners was recorded (D). SNR was determined using the following formula 0.66 x SI1/D * Ö10/3. CNR was determined by the formula SI1-SI2/D. Mean SNR and mean CNR of the 3T MRI group versus 1.5T MRI group were compared. P-values were calculated using paired t-test.  

For each patient, anterior pelvic wall fat thickness (FT) was measured on the midgland slice in the anteroposterior direction. Correlation between SNR and FT and between CNR and FT was calculated using Pearson's correlation coefficient (r) for each system.