Abstract
Prostate cancer is one of the most common forms of cancer found in males making early diagnosis important. Magnetic resonance imaging (MRI) has been useful in visualizing and localizing tumor candidates and with the use of endorectal coils (ERC), the signal-to-noise ratio (SNR) can be improved. The coils introduce intensity inhomogeneities and the surface coil intensity correction built into MRI scanners is used to reduce these inhomogeneities. However, the correction typically performed at the MRI scanner level leads to noise amplification and noise level variations. In this study, we introduce a new Monte Carlo-based noise compensation approach for coil intensity corrected endorectal MRI which allows for effective noise compensation and preservation of details within the prostate. The approach accounts for the ERC SNR profile via a spatially-adaptive noise model for correcting non-stationary noise variations. Such a method is useful particularly for improving the image quality of coil intensity corrected endorectal MRI data performed at the MRI scanner level and when the original raw data is not available. SNR and contrast-to-noise ratio (CNR) analysis in patient experiments demonstrate an average improvement of 11.7 dB and 11.2 dB respectively over uncorrected endorectal MRI, and provides strong performance when compared to existing approaches. A new noise compensation method was developed for the purpose of improving the quality of coil intensity corrected endorectal MRI data performed at the MRI scanner level. We illustrate that promising noise compensation performance can be achieved for the proposed approach, which is particularly important for processing coil intensity corrected endorectal MRI data performed at the MRI scanner level and when the original raw data is not available.
Introduction
Prostate cancer diagnosis relies heavily on high-quality imaging. To achieve the necessary clarity, radiologists often utilize endorectal coils (ERC)—small sensors placed in close proximity to the gland. While these coils significantly boost the signal-to-noise ratio (SNR) locally, they introduce "intensity inhomogeneities," where the image appears excessively bright near the coil and dim further away.
Modern MRI scanners employ built-in software to "flatten" these intensities. However, this correction is a double-edged sword: as it boosts the signal in dim regions, it simultaneously amplifies background noise. This results in non-stationary noise, where image graininess varies across the frame. Standard denoising tools often fail here because they assume uniform noise. This paper introduces Adaptive Coil Enhancement Reconstruction (ACER), a Monte Carlo-based method designed to address these challenges, especially when raw data is unavailable for retrospective studies.
Methods
The ACER framework treats denoising as an inverse problem using Stochastic Bayesian Estimation. Key components include:
- Spatially-Adaptive Monte Carlo Sampling: The algorithm uses a Monte Carlo approach to handle complex, non-uniform noise. It samples pixels in a neighborhood and uses an "acceptance probability" based on a Rician noise model to identify samples representative of the underlying tissue structure.
- Physics-Informed SNR Profiles: ACER incorporates the specific physical characteristics of the coil used (e.g., rigid vs. inflatable). By modeling how coil sensitivity drops with distance, the algorithm adapts its noise model spatially.
- Bayesian Least-Squares: The final reconstruction is the mean of the estimated posterior distribution, balancing noise suppression with detail retention.
The study evaluated ACER using a multi-modality prostate training phantom and clinical data from 14 patients (20 slices), including T2-weighted and Diffusion-Weighted Imaging (DWI).
Results and Discussion
ACER was compared against three industry-standard methods: ROVST, ANLM, and LMMSE.
Quantitative and Qualitative Performance
In patient experiments, ACER achieved an average improvement of 11.7 dB in SNR and 11.2 dB in Contrast-to-Noise Ratio (CNR) over uncorrected images. Blind subjective scoring by seven experts ranked ACER highest for contrast and noise reduction. While some methods like ANLM showed strong edge preservation, they often failed to remove sufficient noise, whereas ACER provided the most "usable" diagnostic image.
Computational Efficiency
ACER represents a middle ground in computational cost, averaging 284 seconds per slice. While slower than the near-instant LMMSE (0.13s), it is significantly faster than the computationally intensive ANLM (1060s).
Conclusion
The proposed ACER method effectively compensates for non-stationary noise in coil intensity-corrected endorectal MRI. By integrating physics-informed SNR profiles and Monte Carlo sampling, it preserves critical anatomical details necessary for prostate cancer diagnosis. Future work will focus on automating coil profiling, optimizing speed for clinical workflows, and extending the method to multi-sequence data integration.
Summary updated