A Unique & Powerful Denoising Approach that Does Not Hallucinate
[1, 2, 6]


Leverage MP-PCA

MICSI-RMT is the first introduction of the popular MP-PCA algorithm into clinical practice. The method utilizes a fast-search algorithm to detect and discard noise-only components that are defined using the Marchenko-Pastur distribution.

The method exploded in popularity for two reasons, (1) its accuracy in removing noise while preserving the signal and (2) its denoising power.

How MP-PCA Eliminates Noise while Preserving Signal

Marchenko-Pastur Law objectively determines denoising thresholds based on learned noise level estimations, ensuring optimal noise removal while preserving essential image features.[1, 2, 6]

MP-PCA Diagram

Smart Averaging

MP-PCA acts as an intelligent averaging tool that combines the many images from the MRI exam, preserving their unique image properties and discarding noise. The degree of image enhancement scales with the number of images in the dataset.

Validated to Preserve Anatomy

MP-PCA does not hallucinate.[1, 2, 6]
The principal component basis enables precise separation between signal and noise, thereby preserving anatomy during denoising.

Physics-based Noise Model

The performance of the algorithm does not depend on externally trained datasets. This approach is self-supervised, meaning that only the images from exam would be used for training and denoising.


Greater Accuracy

All other denoising approaches—filters, wavelets, or even the latest AI neural networks—work by executing weighted combinations of various regions within the image. This process essentially disperses the signal across different parts of the image, leading to loss of signal in addition to noise.

MP-PCA uses the different images over the 4D stack to learn the noise level and remove the components related to noise. If the algorithm fails to detect the noise components, then no denoising occurs, thereby ensuring that this algorithm removes noise without eliminating vital anatomy.

Denoising Methods Applied to a Neuroimaging Diffusion Dataset

The comparison shows the original, MP-PCA, 3-D Gaussian Filter, Anatomical Non-Local Means (ANLM), and a convolutional neural network trained for denoising in 2D. The residuals are plotted for each method to highlight which regions were altered with each denoising method.

MP-PCA is the only method that does not remove anatomy in the brain.[1, 2, 6]





Methods Applied - RMT Diagram
Methods Applied - Gaussian Diagram
Methods Applied - ANLM Diagram
Methods Applied - Neural Diagram


Methods Applied - RMT Residual Diagram
Methods Applied - Gaussian Residual Diagram
Methods Applied - ANLM Residual Diagram
Methods Applied - Neural Residual Diagram

Altered Regions


Denoising Power

MP-PCA can be conceptualized as a smart averaging approach. It combines images with different contrasts, time-points, orientations, or motion states to eliminate noise, as if these images were averaged together. This eliminates the need for in-line averaging and encourages the acquisition of more measurements to increase the information content of the exam. This method fosters a new approach to imaging: acquire as many images with different contrasts as possible (regardless of their SNR), then use them in combination to achieve a significant boost in SNR.

MP-PCA adds a new term to the SNR equation for MRI measurements. This term enables a multiplicative boost in SNR-based on the number of measurements across the entire image series.

Equation Diagram

Enabling New MRI Contrasts

MP-PCA revolutionizes diffusion MRI by enabling high diffusion weightings, or b-values. In high b-value images, the signal throughout the brain is exponentially suppressed, leading to incredibly noisy original images. The suppression is greater for structures that have less restrictions to diffusion of water, such as signals from cerebrospinal fluid and extra-axonal spaces. With MP-PCA it is possible to produce clear, crisp images at high diffusion weightings, unlocking insights into the intricate properties of the brain's intra-axonal regions.



MRI Contrast Original 1 Diagram
MRI Contrast RMT 1 Diagram
MRI Contrast Original 2 Diagram
MRI Contrast RMT 2 Diagram
MRI Contrast Original 3 Diagram
MRI Contrast RMT 3 Diagram
MRI Contrast Original 4 Diagram
MRI Contrast RMT 4 Diagram
MRI Contrast Original 5 Diagram
MRI Contrast RMT 5 Diagram
MRI Contrast Original 6 Diagram
MRI Contrast RMT 6 Diagram
MRI Contrast Original 7 Diagram
MRI Contrast RMT 7 Diagram
MRI Contrast Original 8 Diagram
MRI Contrast RMT 8 Diagram

B = 0

0 5000

Clinical Research

Noise Reduction

Denoising of Diffusion MRI Using Random Matrix Theory...view

Most Cited Paper on Diffusion MRI
1350+ citations from 20+ countries

MP-PCA Enables High Resolution on Low End Scanners

The same patient was imaged with different scanners.

MP-PCA enables high image quality at all resolutions and bridges the gap in image quality between low-end and high-end systems.



High Resolution 1.5 Diagram
High Resolution 3.0 Diagram


High Resolution 1.5 RMT Diagram
High Resolution 3.0 RMT Diagram


Increased Sensitivity with Few Images

Improved Task-based Functional MRI Language Mapping in Patients with Brain Tumors through Marchenko-Pastur Principal Component Analysis Denoising...view

Clinical Sensitivity with Fewer Images

Using MP-PCA, patients reach the same sensitivity level (number of voxels with z > 3) in just 60% of the original imaging time for finger- tapping & verb-generation tasks.


Verb Generation

Verb Generation Task Diagram

Finger Tapping

Finger Tapping Task Diagram

Related Publications

Initial publication describing the method:

Denoising of Diffusion MRI Using Random Matrix Theory

Denoising functional MRI for pre-operative planning:

Improved Task-based Functional MRI Language Mapping in Patients with Brain Tumors through Marchenko-Pastur Principal Component Analysis Denoising

Denoising prostate diffusion MRI on 0.55T:

Feasibility of Accelerated Prostate Diffusion-Weighted Imaging on 0.55 T MRI Enabled With Random Matrix Theory Denoising

Review paper on denoising diffusion MRI data using MPPCA methodology:

Denoising Diffusion MRI: Considerations and Implications for Analysis

Interested in learning more?

Schedule a demo or contact us for more information