Joint estimation of relaxation and diffusion tissue parameters for prostate cancer grading with relaxation-VERDICT MRI

Palombo, Marco ORCID: https://orcid.org/0000-0003-4892-7967, Valindria, Vanya, Singh, Saurabh, Chiou, Eleni, Giganti, Francesco, Pye, Hayley, Whitaker, Hayley C., Atkinson, David, Punwani, Shonit, Alexander, Daniel C. and Panagiotaki, Eleftheria 2021. Joint estimation of relaxation and diffusion tissue parameters for prostate cancer grading with relaxation-VERDICT MRI. [Online]. medRxiv: Cold Spring Harbor Laboratory. Available at: https://doi.org/10.1101/2021.06.24.21259440

Full text not available from this repository.

Abstract

Purpose The non-invasive VERDICT MRI technique has shown promising results in clinical settings discriminating normal from malignant prostate cancer (PCa) tissue and Gleason grade 3+3 from 3+4. However, VERDICT currently doesn’t account for the inherent relaxation properties of the tissue, whose quantification could add complementary information and enhance its diagnostic power. The aim of this work is to introduce relaxation-VERDICT (rVERDICT) for prostate, a model for the joint estimation of diffusion and relaxation parameters from a VERDICT MRI acquisition; and to evaluate its repeatability and diagnostic utility for differentiating Gleason grades. Methods 72 men were recruited and underwent multiparametric MRI (mp-MRI) and VERDICT MRI. Deep neural network was used for ultra-fast fitting of the rVERDICT parameters. 44 men underwent targeted biopsy, which enabled assessment of rVERDICT parameters in differentiating Gleason grades measured with accuracy, F1-score and Cohen’s kappa of a convolutional neural network classifier. To assess repeatability, five men were imaged twice. Results the rVERDICT intracellular volume fraction fic discriminated between 5-class Gleason grades with {accuracy,F1-score,kappa}={8,7,3} percentage points higher than classic VERDICT, and {12,13,24} percentage points higher than the ADC from mp-MRI. Repeatability of rVERDICT parameters was high (R2=0.74–0.99, CV=1%–10%, ICC=78%-98%). T2 values estimated with rVERDICT were not significantly different from those estimated with an independent multi-TE acquisition (p>0.05). The deep neural network fitting provided ultra-fast (∼25x faster than classic VERDICT) and stable fitting of all the rVERDICT parameters.

Item Type:	Website Content
Date Type:	Publication
Status:	Published
Schools:	Psychology
Publisher:	Cold Spring Harbor Laboratory
Last Modified:	10 Nov 2022 10:42
URI:	https://orca.cardiff.ac.uk/id/eprint/147886

Actions (repository staff only)

Edit Item