Assessing Histology Structures by Ex Vivo MR Microscopy and Exploring the Link Between MRM-Derived Radiomic Features and Histopathology in Ovarian Cancer

Abstract

The value of MR radiomic features at a microscopic scale has not been explored in ovarian cancer. The objective of this study was to probe the associations of MR microscopy (MRM) images and MRM-derived radiomic maps with histopathology in high-grade serous ovarian cancer (HGSOC). Nine peritoneal implants from 9 patients with HGSOC were imaged ex vivo with MRM using a 9.4-T MR scanner. All MRM images and computed pixel-wise radiomics maps were correlated with the slice-matched stroma and tumor proportion maps derived from whole histopathologic slide images (WHSI) of corresponding peritoneal implants. Automated MRM-derived segmentation maps of tumor and stroma were constructed using holdout test data and validated against the histopathologic gold standard. Excellent correlation between MRM images and WHSI was observed (Dice index = 0.77). Entropy, correlation, difference entropy, and sum entropy radiomic features were positively associated with high stromal proportion (r = 0.97,0.88, 0.81, and 0.96 respectively, p < 0.05). MR signal intensity, energy, homogeneity, auto correlation, difference variance, and sum average were negatively associated with low stromal proportion (r = -0.91, -0.93, -0.94, -0.9, -0.89, -0.89, respectively, p < 0.05). Using the automated model, MRM predicted stromal proportion with an accuracy ranging from 61.4% to 71.9%. In this hypothesis-generating study, we showed that it is feasible to resolve histologic structures in HGSOC using ex vivo MRM at 9.4 T and radiomics.

Keywords: MRI; histology; machine learning; ovarian cancer; radiomics.

Conflict of interest statement

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Copyright © 2022 Tardieu, Lakhman, Khellaf, Cardoso, Sgarbura, Colombo, Crispin-Ortuzar, Sala, Goze-Bac and Nougaret.

Figures

Figure 1

Illustration of the study experimental workflow.

Figure 2

H&E-stained histological images (A), left) with corresponding high-resolution MR images (A), right), for 4 of the 9 resected peritoneal implants (i–iv). Magnified regions (B) of histological (left) and high-resolution MR (right) images, indicating by red and green boxes on (A). Scale bar at top right of histological images (A) is 1 mm. White and blue arrows on the histologically magnified region (B). (iv) Indicated respectively psammoma bodies and hyaline stroma.

Figure 3

Spatial overlaps between histological images (A) and MR images (B) for stroma and tumor portion. ROIs were manually drawn on MRM and whole slide images and then superposed in (C). Dice similarity index (DSI), stroma proportion, and signal intensity are presented in a table (D) for each ROI.

Figure 4

High-resolution MR image (A) of resected peritoneal implant with corresponding tissue segmentation map (B) and stromal proportion map (C), in %). Corresponding texture maps (D) for energy (i), entropy (ii), and homogeneity (iii) features. For this implant, high tumor proportion (low proportion of stroma) was associated with higher energy (from 0.029 ± 0.020 to 0.019 ± 0.014, t-test p = 0.0002), homogeneity (from 0.44 ± 0.10 to 0.38 ± 0.10, t-test p = 0.0001) and signal intensity (from 141.41 ± 28.30 to 116.09 ± 41.52, t-test p = 0.0002) scores (Figure 3). In contrast, high stromal proportion (low tumor proportion) was associated with higher entropy score (from 5.87 ± 1.00 to 6.47 ± 0.83, t-test p = 0.0002).

Figure 5

Pearson’s correlation plots between texture features and proportion of stroma (%), extracted from resected peritoneal implant from Figure 2. Stromal proportion map was divided into increments of 10 percentage points and mean texture values were calculated for these 10 regions. r indicates the correlation coefficient.

Figure 6

Stroma-rich (>50%) and -poor ((A) and predicted segmentation map from MR feature maps (B), with pixels in green where stromal proportion >50% (stroma-rich) and in red where stromal proportion <50% (stroma-poor). (C) Evaluation map with pixels correctly classified in yellow and incorrectly classified in blue. For four tumors from (i–iv).