Imaging provides valuable non-invasive information for diagnosis and treatment of prostate cancer. The current state of the art is multi-parametric MRI (mp-MRI) (1), although even this lacks specificity and cannot provide reliable grading information. Advanced techniques (2, 3) show promise for probing cancer microstructure and may be more specific than conventional methods. However, rigorous validation is needed to assess the current and potential value of such techniques in prostate cancer management. The current gold standard for validation is histological assessment of whole mount serial sectioned radical prostatectomy specimens, but comparing information from such disparate images presents several challenges.

Below we give an overview of these challenges and the current methods for addressing some of them. Then, we describe a mold-based apparatus and imaging protocol that include ex vivo imaging and a number of innovations that improve alignment of imaging and histology planes and minimize in-plane rotational differences to improve the quality of the 2D registration processes. The apparatus also allows for the collection of supplementary ex vivo MR data for both the fresh unfixed and the fixed prostate specimen.

There are a number of major technical problems complicating any attempt to directly match and correlate MRI and histology data (4). These difficulties include

Unmatched tissue planes. In practice there is usually no direct coordination of presurgical imaging methods and surgical specimen processing. This results in MRI slice planes usually having different plane orientation and plane spacing from the histology images with no clear correspondence between the two so that, at best, only a qualitative correlation of MRI and pathology data is feasible.

The validation problems mentioned earlier have been addressed with varying degrees of completeness by various methods, some of which have been reviewed in Ref. (4, 6). Here, we provide a brief overview of the approaches and problems addressed by the method we describe.

A partial 3D histology “map” can be reconstructed from a stack of histology slices and registered with MR images by a 3D process (7, 8); however, the accuracy of this approach is severely limited by the low out of plane resolution of the MRI data. The problem of missing inter-plane histology data can be addressed by manual (9, 10) or automatic (11) selection of the most closely aligned histology and image planes. Ideally, this reduces the problematic 3D registration to a more tractable 2D process. As the accuracy of this approach is highly dependent on co-alignment of imaging planes with histology sections, a number of methods have been described based on production of a patient-specific 3D-printed mold, the shape of which is defined by the in vivo imaging data. The aim of the mold is to hold the prostate in the same conformation in which it was imaged while guides in the mold align the tissue cutting planes with the image slice positions (12–15). The precision of these mold-based methods is dependent on the degree to which the mold matches the shape of the excised prostate specimen and the amount of any mispositioning or rotation of the prostate in the mold. Rotation errors are more likely around the axes in which the specimen has greatest rotational symmetry. Inclusion of a urethral catheter in the specimen and mold design can decrease rotation errors about the left–right axes of the prostate (6).

High spatial resolution ex vivo imaging of the prostate specimen has been used as an intermediate “stepping-stone” to improve the accuracy of co-registration of in vivo MRI and histology data, both with (6) and without (16) a patient-specific mold. However, the mold design can be improved to better facilitate comparison between MR and histological images.

The following points outline the method of imaging and construction of a patient-specific 3D-printed mold to optimize physical location of the prostate in in vivo MR images, ex vivo MR images, and histology images. The aim is to reduce the inherent 3-dimensional co-registration problem to a more tractable and precise 2-dimensional process.

For ex vivo imaging (both before and after formalin fixation), the specimen in the mold is constrained inside a plastic canister. A liquid-tight seal is made with a piston that also serves as the mounting base for attachment of the sample to the scanner’s small animal bed. Details are described in Figure 3 and below:

Canister. The canister is 68 mm outside diameter, 63.5 mm internal, to fit with 2 mm clearance inside the 72 mm internal diameter RF coil used for 9.4-T ex vivo imaging. The canister is printed in rigid photopolymer (Objet, Stratasys), which is non-porous. Landmarks on the external surface of the canister define the position of the reference slice and top midline of the mold. An internal rib fits the groove in the top of the mold and prevents any rotation of the mold inside the canister, thus ensuring minimal axial plane rotational differences between in vivo and ex vivo images (see Figure 2).

The porous mold design enables fixation of the prostate in situ. The mold/prostate is removed from the canister and immersed in 1-L fixative solution in a narrow vessel placed on a magnetic stirrer overnight. The low density of the nylon mold provides sufficient buoyancy for the prostate to float just below the liquid surface. Use of the stirrer enhances penetration of the fixative through the mold structure and into the prostate. Post fixation, the immersion solution is replaced with 1-L saline for 8 h to dilute the formalin which, at full strength, severely reduces the sample T2 leading to poor signal quality.

Formalin fixation leads to tissue shrinkage (17) and shape changes (18). Fixation of the prostate in the mold constrains any shape changes.

Upon completion of fixation and ex vivo imaging, the prostate is sectioned for histological processing (Figure 4). The mold defines two cutting planes spaced 5 mm apart on either side of the imaging reference plane. Prior to cutting, the mold stabilizer braces (Figure 2) are removed with a scalpel blade, and the mold placed in a dedicated cradle (C). The two cuts on either side of the reference slice are made simultaneously using a pair of skin graft blades (R55170, Rocket Medical, Washington, UK) mounted in an SLS-printed handle (H). The handle and a detachable spacer (S) maintain the blades at 5-mm spacing that aids insertion of the blades into the slots in the top of the mold.

After cutting of the reference plane section, the mold is opened and the prostate removed. The base and apex volumes are then progressively sectioned at 5-mm thickness using a flat plate with 5-mm high rails to guide the blade.

Whole mount tissue sections are processed according to standard laboratory protocol. These sections are cut from the prostate apex face of each block. Note that the reference plane landmarks in the mold are offset toward the apex to account for the histology slice position typically being closer to the apex side of each tissue section than to the base side.

There is a very wide range of 2D image registration methods that could be applied for alignment of the MRI and histology images. As the focus of this paper is on the apparatus used for prostate imaging and sectioning, we present here only an outline of the methods used to produce the example data presented in the figures.

Hematoxylin and eosin stained whole mount sections were digitally scanned with a 20× objective (Hamamatsu NanoZoomer). Histological images were downsampled to 0.25× and converted to grayscale for registration. The high-resolution ex vivo T2 MRI slice for registration was selected based on the mold landmarks. Registration to the corresponding histological slice used an intensity-based 2D rigid registration based on a block-matching strategy (19, 20) with the correlation coefficient as similarity measure. The ex vivo MRI was registered to the in vivo MRI using 2D affine registration, restricting the block matching in the in vivo image to the prostate region using the contour from section 4.2.

An example case is shown in Figure 5.

Registered MRI and histology slices are shown in Figure 6. Slicing artifacts, particularly near the urethra, are visible and may be corrected by additional non-rigid registration techniques. Nevertheless, the spatial correspondence between the ex vivo and histology images can be seen, including in the peripheral zone shape and in the outline of glandular regions within the transition zone. The peripheral zone partially collapses between in vivo and ex vivo imaging, but internal structures within the transition zone show correspondence.

The apparatus detailed here addresses many of the issues described in section 2 and extends previous methods by improving the alignment of in vivo and ex vivo imaging planes with each other and with histopathology sections. Nevertheless, a number of problems remain

Prostate contouring from in vivo images. Contour delineation on the in vivo images aims to trace the expected surgical margins. These are normally close to the prostatic capsule but may vary case by case and the capsule is absent around the prostatic apex. Depending on the T2 image quality and characteristics of the individual prostate, it is sometimes difficult to discriminate the capsule from adjacent intra and extra-prostatic structures. One strategy to accommodate these uncertainties is to print multiple versions of the mold with +10 and −10% adjustments to the axial plane scale and to fit the resected specimen to the mold that provides the best fit.

Apparatus design: RB. Specimen preparation and handling: GS, AS, CB, SH, HP, and AF. Imaging: EJ, CB, and BS. Image registration: TM and CB. Project management: DA, DH, SP, and EP. Manuscript preparation: all.

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.