Exploring the Relationship between MR-derived Apparent Diffusion Coefficient, Cellularity, and Extracellular Porosity:  A Preliminary Animal Study in Prostate Cancer
Deborah K. Hill1,2, Andreas Heindl3, Daniel N. Rodrigues3, Øystein Størkersen2, Yinyin Yuan3, Siver A. Moestue 1,2, Martin O. Leach3, Tone F. Bathen1, David J. Collins3, and Matthew D. Blackledge3

1Norwegian University of Science and Technology, Trondheim, Norway, 2St. Olavs University Hospital, Trondheim, Norway, 3The Institute of Cancer Research and Royal Marsden NHS Foundation Trust, London, United Kingdom

Synopsis

An increased ADC can imply reduced cellularity; DWI is considered a useful tool for assessing tumour treatment response, although there is little validation of this relationship in cancer. We compared ADC, cellularity, and extracellular porosity using a transgenic adenocarcinoma of the mouse prostate model. ADC values were derived from DWI data, and cellularity was assessed from histology using novel visualisation and segmentation tools. We investigated the relationship between extracellular porosity and ADC, and validated our findings using cell segmentation analysis of histology slides. This analysis is useful to inform on tissue cellularity for cases where histology samples are not available.

Background

Apparent diffusion coefficients (ADCs) of tissues, derived from diffusion-weighted MRI (DWI), are often linked with tissue cellularity (number of cells per unit volume). In cancer it is normally hypothesised that an increase in ADC gives evidence of reduced cellularity, so DWI is considered a useful tool for assessing tumour response to treatment [1], although there is still little validation of this relationship in cancer. In this study ADC, cellularity, and extracellular porosity were compared using a transgenic adenocarcinoma of the mouse prostate (TRAMP) model. ADC values were derived from DWI data, and cellularity scores were assessed from histology using novel visualisation and segmentation tools: Polyzoomer [2] and CRImage [3]. Further, we investigated the relationship between extracellular porosity and ADC, and validated our findings using cell segmentation analysis of histology slides. This analysis may be useful to inform on tissue cellularity for cases where histology samples are not available.

Methods

Animals: An initial study on TRAMP (n=3) and control (c57BL/6) (n=1) mice, taken from a larger cohort, were imaged using MRI every 4 weeks from 8 weeks of age. TRAMP mice were monitored for cancer onset using high-resolution T2-weighted (HR T2W, in plane resolution 0.1x0.1mm2) images and ADC maps obtained from DWI. Mice were terminated between 28-30 weeks of age or when visual inspection of images indicated unacceptable tumour burden. Upon sacrifice, the GU tract (prostate, seminal vesicles, bladder) was excised and formalin-fixed for histological analysis (hematoxylin, eosin and saffron staining (HES)). TRAMP prostates were classified according to the histology as either well-differentiated adenocarcinoma or poorly differentiated carcinoma.

Imaging: MRI was performed on a 7T Bruker Biospec magnet with an 86mm diameter volume resonator (transmission) and a surface coil for reception. DWI was performed using a multi-shot EPI sequence: TE=28.5ms, TR=3000ms, averages=4, matrix size=128x128, slice thickness=1.0mm, in plane resolution 0.2x0.2mm2, and b-values=0, 100, 200, 400, 800 s/mm2 along three orthogonal gradient orientations. ADC estimation was performed using least-squares, mono-exponential fitting.

Histological Analysis: Formalin fixed paraffin embedded samples were sectioned (slice thickness: 4μm) and HES stained. Slides were digitized using a Hamamatsu NanoZoomer XR (Hamamatsu, Japan) scanner (40x) and visualised using Polyzoomer (ICR, London, UK) (Fig. 1). Each pixel in the high-resolution images represents 0.23µm. To compute cellularity maps, HES stained sections were analysed employing CRImage; the classifier was trained to define cellular components as nucleus or artefact. Cellularity, C, was estimated by finding the total number of nuclei in each pixel and dividing by the pixel area (Fig. 2). Extracellular tissue porosity, ε, was determined from the cellularity by assuming a spherical cell size of fixed radius, r:

ε = 1-VC/VP ≈ 1-⅔Ÿ·C·(2πr2) (1)

where VC is the total cell volume within the pixel volume, VP, r is the mean radius calculated from multiple measurements from the HES samples (Fig. 3).

Statistical Analysis: Regions of Interest (ROIs) differentiating between ventral prostate, dorsal prostate, and cancer (where applicable) were drawn on the cellularity maps using histology images as reference. Corresponding ROIs were drawn on b0 DW-MR images and transferred to ADC maps. Mean and standard deviation of cellularity and ADC derived from each region were compared on a scatter diagram (Fig. 4). By equating the model of tortuosity derived for ADC measurements [4] with Archie’s law for porous media [5] we fit the following model using a linear least-squares approach:

ln(ε) = (1/2n)Ÿln(ADC) – (1/2n)Ÿln(Dfree) (2)

where Dfree is the diffusion coefficient of free water at body temperature and n is a scaling constant.

Results and Discussion

Good registration was attained between MRI and histology (Fig. 1). CRImage successfully delineated nuclei (Fig. 2, Polyzoomed image available [6]). A plot of cellularity against ADC revealed a clustering of different tissue types (Fig. 4); the relationship was investigated further by plotting ln(ADC) against ln(extracellular porosity) (Fig. 5) which showed an agreement between the model prediction and experimental data. The fit of model (2) provided an estimate for the scaling term of n = 3.69 ± 0.16, which in turn was used to estimate Dfree = 2.7x10-3 mm2/s, which is within the range expected by literature [7]. An additional consideration, however, is that we have not accounted for the confounding effects of water exchange or flow on ADC estimates.

Conclusion

Using CRImage for automated segmentation and estimation of nuclear count demonstrated a relationship between ADC value and cellularity in prostates from TRAMP and control mice. Our approach suggests that DWI can inform on cellularity in cases where it is not feasible to acquire tissue for histological examination.

Acknowledgements

We acknowledge support from the liaison Committee between the Central Norway Regional Health Authority and the Norwegian University of Science and Technology, CRUK in association with MRC and Department of Health, NHS funding to the NIHR Biomedicine Research Centre, and post-doctoral fellowship funding by the NIHR. MRI was performed at the MR Core Facility, NTNU, animals were housed at the Comparative Medicine Core Facility, NTNU, and histology performed at the Cellular and Molecular Imaging Core Facility, NTNU.

References

[1] Padhani A. et al. JMRI 2014, [2] Polyzoomer is developed by YuanLab, http://www.yuanlab.org, [3] Yuan Y. et al. Sci Transl Med. 2012, [4] Khanafer, K. and Vafai, K., Heat Mass Trans., 2006, [5] Pfeuffer, J. et al. Mag Reson. Imag. 1998., [6] http://bit.ly/1wZgW6y, [7] Holz, M. et al. Phys. Chem. Chem. Phys., 2000.

Figures

Figure 1: (A) HR-T2W MRI of a TRAMP prostate. Ventral prostate (VP), dorso-lateral prostate (DP), seminal vesicle (S.V.), well-differentiated cancer (red arrow) is indicated in each of the images. (B) HES stained histology slide corresponding with (A). (C) Registered ADC map.

Figure 2: Segmentation of nuclei using CRImage; example from a poorly differentiated cancer region. Nuclei are outlined in white. (Polyzoomed image: [6]). The number of counted nuclei per unit area is converted into an estimated cellularity map.

Figure 3: We convert H&E stained slides into the LAB color space and use the lightness channel to visualise cell nuclei. Multiple ROIs were drawn to estimate the cell radius and results plotted as a histogram (right). The mean cell radius in this case was found to be 4.02±0.27 μm.

Figure 4: Scatter plot of mean cellularity against mean ADC from ROIs. Data are mean±S.E. of ROI values. Cancer (red), TRAMP prostate tissue (black) and c57BL/6 control prostate tissue (green) are clustered.

Figure 5: A plot of the logarithm of estimated ADC against estimates of extracellular porosity reveals a linear trend as predicted by model (1). Percentage confidence intervals (%CI) for the line of best fit (red dashed) are displayed on a color scale.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
0444