Simplified Luminal Water Imaging for the Detection of Prostate Cancer From Multiecho T2 MR Images

William Devine, Francesco Giganti, Edward W. Johnston, Harbir Sidhu, Eleftheria Panagiotaki, Shonit Punwani, Daniel C. Alexander, David AtkinsonJournal of Magnetic Resonance Imaging 2018 · DOI

TL;DR

This paper introduces a simplified Luminal Water Imaging (LWI) technique that shows comparable accuracy to diffusion-weighted imaging (DWI) for detecting prostate cancer, but without the common artifacts associated with DWI. This novel approach offers a promising alternative for improving prostate cancer diagnosis using multiparametric MRI.

What Problem Does This Paper Address?

Prostate cancer detection often relies on multiparametric MRI (mp-MRI), which typically includes T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and dynamic contrast-enhanced (DCE) MRI. These images are then scored using the PI-RADS v2 system. However, current mp-MRI methods face challenges: DWI sequences can suffer from distortion and signal artifacts, leading to less accurate results and a significant proportion of indeterminate lesions. While DWI is a key component of PI-RADS v2 scoring, its limitations necessitate alternative or complementary techniques. This paper aims to address these limitations by evaluating a simplified LWI technique as a potential replacement or adjunct to DWI for prostate cancer detection.

Methods

The researchers developed a simplified Luminal Water Imaging (LWI) technique that models the prostate tissue using two compartments: a luminal compartment with long T2 values and a stromal/epithelial compartment with shorter T2 values. This model is fitted to data acquired from multiecho T2 (ME-T2) MRI sequences. The key innovation lies in simplifying the fitting process by constraining the model to two Gaussian distributions for T2 values, reducing the number of parameters compared to previous LWI methods. This simplified approach aims to improve reliability and reduce data requirements.

For comparison, the study also utilized apparent diffusion coefficient (ADC) maps derived from DWI sequences. Regions of interest (ROIs) were manually drawn by a radiologist on both ME-T2 and ADC images, corresponding to areas of lesion and benign tissue, and assigned PI-RADS v2 scores. Statistical analyses, including receiver operating characteristic (ROC) and logistic regression, were employed to compare the performance of LWI and ADC in classifying lesions based on PI-RADS v2 scores and in differentiating histologically confirmed malignant from nonmalignant tissue.

Datasets

The study utilized in vivo data from a subset of 65 male patients aged 49-79 years who were referred for prostate mp-MRI. These patients had either undergone a previous biopsy more than 6 months prior or presented with clinical suspicion of prostate cancer. Histological grading (Gleason score) was available for a subset of 31 ROIs. The study also included simulation data to evaluate the LWI fitting method.

Key Insights and Ideas

  • LWI as a Viable Alternative to DWI: The study demonstrates that LWI can achieve comparable diagnostic performance to ADC (derived from DWI) in classifying prostate cancer lesions according to PI-RADS v2 scores, despite DWI being a cornerstone of the PI-RADS system.
  • Artifact Reduction: A significant advantage of LWI highlighted is its reduced susceptibility to the distortion and signal pile-up artifacts commonly seen in EPI-based DWI sequences, leading to potentially more reliable imaging.
  • Microstructural Sensitivity: The correlations found between LWI parameters (like LWF) and the VERDICT diffusion model parameters (fIC, fEES) suggest that LWI is sensitive to the underlying tissue microstructure, providing biophysically interpretable information. This is a key advantage over ADC, which offers less direct microstructural insight.
  • Simplified Modeling: The development of a simplified two-Gaussian LWI model proves effective, showing similar accuracy to more complex models while requiring fewer data points and potentially fewer echoes, making it more practical for clinical implementation.

Key Results

  • PI-RADS v2 Classification: LWI achieved an area under the curve (AUC) of 0.779 for classifying tissue into PI-RADS 1,2 vs. PI-RADS 3,4,5 groups, which was comparable to ADC's AUC of 0.764.
  • Malignant vs. Nonmalignant Differentiation: LWI demonstrated superior performance in differentiating histologically confirmed malignant from nonmalignant tissue, with AUC, sensitivity, and specificity values of 0.81, 75%, and 87%, respectively. ADC achieved AUC of 0.75, sensitivity of 83%, and specificity of 67%.
  • Reproducibility: A Bland-Altman analysis on a subset of ROIs showed good reproducibility for the median LWF values (r² = 0.928).
  • Correlation with VERDICT: Significant correlations were observed between LWI parameters and VERDICT model parameters, particularly with fIC (intracellular fraction) and fEES (extracellular-extravascular fraction), supporting LWI's sensitivity to tissue microstructure.

Limitations and Future Work

  • Histological Correlation: A limitation acknowledged is that not all patients underwent biopsy, limiting the ability to perform full histological correlations for all lesions. The study also notes that biopsy is prone to sampling error, which can lead to undergrading of tumors.
  • Retrospective Analysis: The study was retrospective, using a subset of patients from a larger prospective study.
  • Reader Variability: While reproducibility was assessed, the authors suggest that future work should include a multiple-reader study to quantify variability in ROI placement, especially if ROIs are drawn directly on LWF maps.
  • Future Directions: The authors propose that LWI could be a valuable tool for PCa detection, potentially reducing the need for extensive diffusion imaging or complementing it. Future work could involve sequence optimization for wider prostate coverage and thinner slices, as well as multicenter studies to validate these findings on larger cohorts. They also suggest that a reduced number of protocols for grading prostate cancer could be possible with further validation.

Summary updated

Abstract

3 Technical Efficacy: Stage 2 J. Magn. Reson. Imaging 2019;50:910-917.
BibTeX
@article{devine2018simplified,
  author = {William Devine and Francesco Giganti and Edward W. Johnston and Harbir Sidhu and Eleftheria Panagiotaki and Shonit Punwani and Daniel C. Alexander and David Atkinson},
  title = {{Simplified Luminal Water Imaging for the Detection of Prostate Cancer From Multiecho T<sub>2</sub> MR Images}},
  journal = {Journal of Magnetic Resonance Imaging},
  year = {2018},
  doi = {10.1002/jmri.26608},
}
Figure