Prostate Cancer: A Correlative Study of Multiparametric MR Imaging and Digital Histopathology

Jin Tae Kwak, Sandeep Sankineni, Sheng Xu, Barış Türkbey, Peter L. Choyke, Peter A. Pinto, Vanessa Moreno, Maria J. Merino, Bradford J. WoodRadiology 2017 · DOI

TL;DR

This study bridges the gap between non-invasive prostate MRI and detailed digital histopathology (DHA) by correlating imaging features with tissue micro-architectural components on a voxel-by-voxel basis. The findings demonstrate significant associations between MRI signals and tissue densities, offering a path towards more precise characterization of prostate cancer aggressiveness and potentially improving diagnostic accuracy.

What Problem Does This Paper Address?

Prostate cancer diagnosis and grading heavily rely on the Gleason score, a qualitative assessment of tumor histology. While multiparametric MRI (mpMRI) has shown promise in detecting and localizing prostate cancer, its specificity can be limited by benign prostatic conditions that mimic malignant features. Furthermore, the Gleason score, while crucial, doesn't fully capture tumor heterogeneity. Digital histopathology (DHA) offers a quantitative, voxel-level analysis of tissue microstructures (lumen, epithelium, stroma, epithelial nucleus), but its integration with non-invasive imaging has been challenging due to registration complexities and potential biases from manual region selection. This paper aims to address this by establishing a systematic, voxel-wise correlation between mpMRI parameters and DHA, moving beyond region-based analyses to provide a more granular understanding of their relationship.

Methods

The researchers employed a retrospective study design, analyzing data from 40 patients who underwent mpMRI and subsequent robot-assisted radical prostatectomy. A key methodological innovation was the use of a patient-specific mold (PSM) created from presurgical MRI data and fabricated with a 3D printer. This PSM, with precisely positioned sectioning slots, ensured accurate alignment of whole-mount tissue slices with the original MRI data.

After obtaining tissue specimens and performing hematoxylin-eosin staining, DHA was conducted. This involved segmenting the digitized tissue images into four components: lumen, epithelium, epithelial nucleus, and stroma. Machine learning techniques were used for this segmentation. Crucially, image registration algorithms were developed to spatially align the DHA images with the mpMRI data (T2-weighted, diffusion-weighted, and high-b-value diffusion-weighted images).

The core of the analysis involved correlating MR imaging signal intensities and DHA-derived tissue component densities. This was performed using both region-based (averaging within defined regions of interest) and voxel-based (per-voxel analysis) approaches, employing linear mixed-effects models and Pearson correlation coefficients. To account for potential variations in MR signal intensity across different sections and patients, normalization techniques were applied.

Datasets

The study utilized data from 40 patients who underwent mpMRI and robot-assisted radical prostatectomy. This included:

  • Multiparametric MRI data: T2-weighted imaging, diffusion-weighted (DW) imaging (including apparent diffusion coefficient - ADC maps), and high-b-value DW imaging.
  • Digital Histopathology (DHA) data: Whole-mount tissue specimens from these patients, digitized and segmented into lumen, epithelium, epithelial nucleus, and stroma.

The study did not introduce a new benchmark dataset but rather leveraged existing clinical data and developed novel methods for registration and correlation.

Key Insights and Ideas

  • Voxel-wise correlation is key: Moving beyond region-based analysis to a voxel-by-voxel correlation between MRI and DHA is crucial for capturing finer details and understanding tissue heterogeneity.
  • MRI signals reflect tissue composition: Specific MRI sequences (T2W, ADC, high-b-value DW) are not just indicators of tissue presence but are quantitatively linked to the relative proportions of lumen, epithelium, and stroma. This suggests MRI is capturing more than just gross structural changes.
  • Benign vs. Malignant differentiation: Significant differences in both MRI signals and tissue component densities were observed between benign and malignant regions, highlighting the potential for improved specificity in cancer detection.
  • Gleason score linkage: While not a perfect one-to-one mapping, Gleason score showed significant associations with specific MRI and DHA parameters, particularly with high-b-value DW imaging and lumen density. This hints at MRI's ability to indirectly reflect the aggressiveness captured by Gleason grading.
  • Anatomical differences matter: The study observed regional differences in the association between MRI parameters and Gleason score, particularly between the peripheral zone (PZ) and transition zone (TZ), suggesting that a one-size-fits-all approach might not be optimal.

Key Results

  • Significant correlations: T2-weighted MRI, ADC, and high-b-value DW imaging were all significantly related to specific DHA parameters (P < .01).
  • Lumen density associations: Lumen density showed positive correlations with T2-weighted MRI (slope = 0.36) and ADC (slope = 0.47), and a negative correlation with high-b-value DW imaging (slope = -0.44).
  • Benign vs. Malignant differences: Significant differences were observed between benign and malignant regions in both MRI and DHA (P < .01).
  • Gleason score correlations: Gleason score was positively related to high-b-value DW imaging (slope = 0.21) and negatively related to lumen density (slope = -0.19).
  • Regional variations: The study found that T2-weighted MRI was a better indicator for tumors in the TZ, while both T2-weighted and DW imaging (including ADC) were useful for assessing aggressiveness in the PZ.

Limitations and Future Work

The authors acknowledge several limitations:

  • Registration accuracy: While efforts were made to optimize registration, errors could still occur, potentially affecting the strength of associations. The lack of a definitive "ground truth" for registration accuracy made quantitative assessment difficult.
  • Fragmented tissues and artifacts: Tissue preparation and handling can introduce artifacts that interfere with registration and analysis.
  • Limited MRI sequences: The study primarily focused on T2-weighted and DW imaging. Incorporating other sequences like dynamic contrast-enhanced MRI or quantitative T2 maps could provide a more comprehensive picture.
  • ROI selection bias: The inclusion of only ROIs identified by both radiologists and pathologists might introduce bias.
  • Small sample sizes for certain tumor types: The number of tumors with specific Gleason score patterns (e.g., 4+3 or 4+5 in the TZ) was small, limiting generalizability. The absence of Gleason score 3+3 tumors in the analysis is also noted.

Future work suggested by the authors includes: * Investigating texture characteristics: Exploring correlations with texture features in both MRI and tissue specimens could offer further insights. * Incorporating other MRI techniques: Expanding the study to include dynamic contrast-enhanced MRI and quantitative T2 maps. * Larger studies: Conducting larger studies to overcome limitations related to sample size and tumor heterogeneity. * Developing computer-aided design systems: Using the insights gained to train systems that can prospectively identify aggressive tumor regions, leading to more targeted biopsies and improved cancer management. * Further research on stromal contributions: The inconsistent association with stroma warrants further investigation into its subcomponents and their impact on MRI signals.

Summary updated

Abstract

Purpose To correlate multiparametric magnetic resonance (MR) imaging and quantitative digital histopathologic analysis (DHA) of the prostate. Materials and Methods This retrospective study was approved by the local institutional review board and was HIPAA compliant. Forty patients (median age, 60 years; age range, 44-71 years) who underwent prostate MR imaging consisting of T2-weighted and diffusion-weighted (DW) MR imaging along with subsequent robot-assisted radical prostatectomy gave informed consent to be included. Whole-mount tissue specimens were obtained with a patient-specific mold, and DHA was performed to assess the lumen, epithelium, stroma, and epithelial nucleus. These DHA images were registered with MR images and were correlated on a per-voxel basis. The relationship between MR imaging and DHA was assessed by using a linear mixed-effects model and the Pearson correlation coefficient. Results T2-weighted MR imaging, apparent diffusion coefficient (ADC) of DW imaging, and high-b-value DW imaging were significantly related to specific DHA parameters (P < .01). For instance, lumen density (ie, the percentage area of tissue components) was associated with T2-weighted MR imaging (slope = 0.36 ± 0.05 [standard error], γ = 0.35), ADC (slope = 0.47 ± 0.05, γ = 0.50), and high-b-value DW imaging (slope = -0.44 ± 0.05, γ = -0.44). Differences between regions harboring benign tissue and those harboring malignant tissue were observed at MR imaging and DHA (P < .01). Gleason score was significantly associated with MR imaging and DHA parameters (P < .05). For example, it was positively related to high-b-value DW imaging (slope = 0.21 ± 0.16, γ = 0.18) and negatively related to lumen density (slope = -0.19 ± 0.18, γ = -0.35). Conclusion Overall, significant associations were observed between MR imaging and DHA, regardless of prostate anatomy. <sup>©</sup> RSNA, 2017 Online supplemental material is available for this article.
BibTeX
@inproceedings{kwak2017prostate,
  author = {Jin Tae Kwak and Sandeep Sankineni and Sheng Xu and Barış Türkbey and Peter L. Choyke and Peter A. Pinto and Vanessa Moreno and Maria J. Merino and Bradford J. Wood},
  title = {{Prostate Cancer: A Correlative Study of Multiparametric MR Imaging and Digital Histopathology}},
  booktitle = {Radiology},
  year = {2017},
  doi = {10.1148/radiol.2017160906},
}
Figure