Prostate diffusion imaging may become clearer when improved coil design is combined with deep learning reconstruction. A prospective single-centre feasibility analysis published in Radiology Advances assessed a prototype 50-channel pelvic coil used with deep learning denoising and phase correction for prostate multiparametric MRI. The work enrolled 24 consecutive men referred for prostate imaging over 16 months, with 20 included in qualitative assessment and 18 included in quantitative assessment. The strongest performance came from the 50-channel coil combined with deep learning phase correction and AIR Recon DL denoising. This configuration produced the highest image quality, with better signal measures and stronger reader scores for prostate borders, zonal distinction, lesion visibility and confidence in extraprostatic extension.
High-Density Coil Design Addresses Imaging Constraints
Diffusion-weighted imaging plays a central role in prostate multiparametric MRI, especially when assessing areas suspicious for clinically significant cancer. High b-value diffusion images and apparent diffusion coefficient maps support lesion visibility, but image quality can be limited by weak signal, background noise and susceptibility artefacts. These issues may obscure anatomy and make interpretation more difficult.
Must Read: Clinical Data Strengthens Prostate MRI Models
The prototype 50-channel pelvic coil aims to improve image quality by bringing coil elements closer to the prostate. Its flexible design uses a posterior base, two lateral flaps and a centre flap to cover the pelvis and perineal region. The centre flap adds coverage in an area that standard anterior and posterior arrays do not cover in the same way. Routine imaging with a standard 30-channel anterior array and posterior array served as the clinical baseline. The 50-channel coil alone improved perceived signal and offered modest gains in diffusion image quality compared with the standard coil arrangement. However, noise remained substantial at high b-value, limiting the effect of the hardware change when denoising was used without the additional phase correction step.
Deep Learning Reconstruction Strengthens Image Clarity
The reconstruction approach combined AIR Recon DL denoising with a prototype deep learning phase correction method. Standard product reconstruction uses a low-pass filter to correct phase variation before signal averaging. The prototype method replaces that step with a deep learning model designed to generate a higher-quality phase map before averaging. This phase correction step further reduced background noise when applied to the same acquisitions. On high b-value diffusion images, the combined reconstruction improved gland delineation, zonal distinction and lesion conspicuity. Lesion borders appeared more distinct in examples shown with the combined 50-channel coil and deep learning reconstruction pipeline.
The same improvement carried into derived images. Apparent diffusion coefficient maps became more uniform in low-signal areas, including muscle. Synthetic high b-value images also showed clearer prostate borders and better contrast within the peripheral zone when reconstructed with the combined pipeline. These derived images depend on the quality of the underlying diffusion data, so better input images supported clearer output images across the assessed series.
Reader Scores Favour the Combined Approach
Two radiologists independently scored the reconstructed images using a five-point image quality scale. They assessed prostate border visualisation, distinction between the peripheral and transition zones, lesion conspicuity and confidence in extraprostatic extension. T2-weighted images served only as anatomical reference images and were not scored. The combined 50-channel coil and deep learning phase correction pipeline received the highest overall image quality scores. Reader scores improved when moving from the standard coil with denoising to the 50-channel coil with denoising, then improved further when deep learning phase correction was added.
The same pattern appeared for acquired high b-value diffusion images and apparent diffusion coefficient maps. Synthetic high b-value images also improved, particularly for border visualisation, zonal distinction and confidence in extraprostatic extension. Quantitative results supported the visual assessment. The 50-channel coil improved signal-to-noise and contrast-to-noise ratios compared with the standard coil configuration, while the combined 50-channel and phase correction approach achieved the highest values. Agreement between the two readers was fair for acquired diffusion images and lower for synthetic images, although both readers showed the same overall direction of preference.
The combination of a prototype 50-channel pelvic coil, deep learning phase correction and denoising produced the clearest prostate diffusion images among the tested configurations. The approach improved signal measures, reduced background noise and strengthened visual assessment of prostate anatomy and suspicious lesions. These gains extended to apparent diffusion coefficient maps and synthetic high b-value images. The results focus on image quality rather than diagnostic accuracy. Further evaluation is needed to establish diagnostic value, workflow impact and performance in larger clinical settings.
Source: Radiology Advances
Image Credit: iStock
References:
Huang SS, Wang X, Lan P et al. (2026) Improved prostate diffusion imaging using deep learning denoising and phase correction with ultra-high-density coil array. Radiology Advances, 3(2): umag019.
