Postmortem Magnetic Resonance Imaging to Guide the Pathological Cut: Individualized, 3D-Printed Cutting Boxes for Fixed Brains

Abstract

Interfacing magnetic resonance imaging (MRI) and pathology is critically important for understanding the pathological basis of MRI signal changes in vivo and for clinicopathological correlations. Postmortem MRI is an intermediate step in this process; unfortunately, however, relating the data to standard pathological sections, which are relatively thick and often non-parallel, is both time consuming and insufficiently accurate. The aim of this project was to develop technology to integrate postmortem, high-resolution, whole-brain MRI into the planning and execution of the pathological analysis through precise localization of the target and coordinates of cut. Compared to standard pathological sectioning, the use of an individualized 3D-printed cutting-box, designed based on postmortem MRI of formalin-fixed whole brains, improved the speed, quality, and accuracy of radiological-pathological correlation and, specifically, the histopathological localization of imaging findings. The technology described herein is easily implemented, applicable to any brain disorder, and potentially extendable to other organs. From the point of view of the pathologist this technique can improve localization of small or subtle abnormalities, whereas from the point of view of the radiologist it has the potential to improve understanding of MRI signal changes observed in disease.

INTRODUCTION

There is a growing need for interfacing brain morpho-pathology with magnetic resonance imaging (MRI) to improve understanding of the pathological basis of MRI signal changes and thereby validate the in vivo surrogates of physiological and pathological processes (1–5). Postmortem MRI is a valuable intermediate step to this end. In particular, it as a bridge from in vivo images to pathological examination. In this multimodal context, pioneering experience in combining digitalized whole-mount histology and MRI has been recently realized for the whole brain of a famous patient with epilepsy (6), suggesting that MRI might constitute the topographic baseline reference for the 3D histopathology of the future (7). In cases where whole-brain histopathology is not required or feasible, in vivo and postmortem MRI can be used as a guide for limited pathological sampling thereby saving valuable resources (8–10). Similarly, in forensic radiology, postmortem MRI has been recognized as a supplementary diagnostic tool that is able not only to provide a general overview of the corpse but also to uncover small findings and address specific forensic questions (11, 12). In this setting, the term “virtual autopsy” has been recently proposed as a topic for future research (13).

At this time, however, the tools that are required for precise colocalization are not routinely implemented in pathological examinations (14). In general, sectioning of the brain, either coronally, sagittally, or horizontally, leads to large variations in slice thickness and in the degree to which the slice faces remain parallel. Additionally, subsequent sampling of small tissue blocks for in-depth histological analysis is frequently blind to findings, particularly small ones that are not visible on the slice surfaces after the initial cut. Integrating and comparing data derived from the standard procedure with in vivo and postmortem MRI is time consuming and often not sufficiently accurate.

The aim of this project was to develop a technology to integrate postmortem, high-resolution, whole-brain MRI into the planning and execution of the pathological analysis through precise localization of the target structure and cutting coordinates. There are several reasons why postmortem MRI acquisition is preferentially performed on the whole fixed brain instead of tissue slices. These include fewer artifacts (air bubbles and distortion at the slice edges), preserved landmarks for radiological interpretation, and reduced total scan time to cover the whole brain. Additionally, whole-brain acquisition allows construction of a customized, 3D-printed, brain-slicing box – the main innovation proposed herein. Based on our results, we propose that in vivo and postmortem high-resolution MRI can help to guide pathology studies, particularly in the research setting.

MATERIALS AND METHODS

Tissue samples

Three autopsy brains were obtained with the consent from the next of kin. Clinical information and in vivo MRI scans were obtained under an IRB-approved clinical research protocol and compared with the postmortem data for Patient 1. Similarly, clinical information and in vivo MRI scans were obtained from cases 2 and 3 as part of clinical-pathological studies under an autopsy protocol at Johns Hopkins Hospital.

Patient 1 was a 59-year-old man with primary progressive multiple sclerosis (MS), with disease duration (i.e. time from symptom onset to death) of 21 years. The cause of death was respiratory distress and sepsis following multiple embolic strokes. Patient 2 was a 55-year-old woman with secondary progressive MS with disease duration of more than 20 years; the cause of death was pneumonia. Patient 3 was a 22-year-old woman with anti-N-methyl D-aspartate (NMDA)-receptor encephalitis that was not responsive to intravenous steroids, plasma-exchange, intravenous immunoglobulin, or rituximab. The disease duration was ~6 months and the cause of death was a sudden cardiac arrest, likely secondary to dysautonomia.

Postmortem to autopsy intervals were, respectively, 5 hours (UPMC Presbyterian, Department of Pathology, Pittsburgh, PA), 28 hours and 7 hours (Johns Hopkins Hospital, Baltimore, MD). The brains were extracted en bloc and subsequently fixed by immersion in 10% formalin for 2 weeks. Following MRI at the NIH, sectioning was performed using a customized cutting box (described below) for Patients 1 and 3 and using standard procedures for Patient 2 (Table).

Table.

Summary of Sectioning and Histological Procedures

Patient No.	Diagnosis	Sectioning Procedure	Histology	MRI comparison
		Forebrain with cutting box (23 6-mm-thick coronal slices)	Block#1 L temporal lobe (1×3 inch glass slides)	Immediate and accurate (block sectioned according to MRI coordinates)
#1	Multiple sclerosis	Brainstem-cerebellum with cutting box (9 6-mm-thick transversal slices)	NA	NA
#2	Multiple sclerosis	Forebrain without cutting box (15 1-cm-thick coronal slices)	Block #2 L frontal lobe (2×3 inch glass slides)	A posteriori, imaging 3D reformatting is necessary
#3	Anti-NMDA-receptor encephalitis	Forebrain with cutting box (21 6-mm-thick coronal slices)	Block #3 L hippocampus (1×3 inch glass slides)	Immediate and accurate (block sectioned according to MRI coordinates)

NMDA, N-Methyl-D-Aspartate; L, left; R, right; NA, not available.

Postmortem 7 tesla MRI

The formalin-fixed forebrain was first separated from the brainstem and cerebellum by cutting along a transversal plane passing though the midbrain at the level of the substantia nigra. The forebrain was then positioned within an imaging container consisting of a cylindrical tube of 25 cm inner diameter, 30-cm length, and 0.5 cm wall thickness (Fig. 1A). The container was fitted with a hemispherical dome at one end for optimal magnetic susceptibility characteristics and a sealable spout at the other for filling and applying suction. The shape of the container was such that it optimally fit the volumetric receive coil used on the 7 tesla MRI scanner; the brain was positioned approximately at the center of the transmit coil. The container was filled with Fomblin (Solvay Solexis, West Deptford, NJ), a perfluropolyether that is susceptibility-matched to tissue, thereby reducing artifacts in the images. Air bubbles were removed through the spout by gentle suctioning for ~30 minutes.

Figure 1.

Illustration of the technique. (A–C) A dome-shaped container was fashioned for magnetic resonance imaging (MRI) of the postmortem brain; it was customized to reduce susceptibility artifacts and to fill a 7 tesla MRI receive coil (A, left panel). The postmortem brain was immersed in Fomblin and aspirated for air bubbles through the sprout by applying suction for ~30 minutes before imaging (A, middle and right panels). For Patient 1, individualized cutting boxes were designed and 3D-printed for the forebrain (B) and the brainstem-cerebellum (C). The surface of the fixed brain was rendered from the MRI sequence, and a mold was created to conform to the inferior surface of the brain (B) and posterior surface of the cerebellum (C).

Postmortem MRI was performed on a 7 tesla actively shielded scanner (Siemens, Erlangen, Germany) equipped with a birdcage-type transmit coil and a 32-channel receive coil. The following pulse sequences were executed:

1. 3D T1-weighted magnetization-prepared rapid gradient echo (T1-MPRAGE) with repetition time (TR) = 2200 ms; echo time (TE) = 3.04 ms; inversion time (TI) = 1050 ms; flip angle (FA) = 7°; nominal resolution = 0.6×0.6×0.6 mm; and 176 coronal slices. The acquisition time for the sequence was 6 minutes 35 seconds.

2. 3D high-resolution multigradient-echo (GRE) T2* sequence with TR = 60 ms; TE = 6.09, 15.99, 25.89, 35.79 ms; 4 averages; 88 slices; FA = 10°; acquisition time = 2 hours 15 min per 36 mm slice; nominal resolution = 0.42×0.42×0.42 mm. Four coronal slices with 20% overlapping slices were acquired to cover the whole brain.

3. 3D high-resolution Fast Low Angle Shot (FLASH) T1-weighted sequence with TR = 1000 ms; TE = 2.71 ms; TI = 150 ms; 4 averages; 192 slices; FA = 10°; acquisition time = 2 hours 45 min per 60 mm slice; nominal resolution = 0.31×0.31×0.31mm. Three coronal slices with 20% overlapping slices were acquired to cover the whole brain.

All imaging slices were aligned parallel to a plane passing through the mammillary bodies. Images acquired in multiple slices covering the whole brain were stitched together using their DICOM information and post-processing algorithms developed using AFNI. The typical scan time for the entire brain was about 17 hours, though only the ~6.5-minute T1-MPRAGE scan was used for preparation of the cutting box.

Individualized cutting box

3D T1-MPRAGE images were initially processed using MIPAV (Medical Image Processing, Analysis & Visualization, NIH, Bethesda, MD) to generate the design for the individualized cutting box for each brain. According to specific findings of interest, the number, the position and the thickness of each slice were determined a priori using MRI images. The surface of the forebrain was rendered and a mold was created to conform to the inferior (skull base) surface (Fig. 1B). To implement this, the coronal 3D T1-MPRAGE (Supplementary Fig. 1A, B, left column) was thresholded to obtain the brain mask (Supplementary Fig. 1A, center). An accurate cleaning of residual voxels belonging to the parenchyma or ventricles was performed manually and took approximately 1 hour to complete. Parallel, coronally oriented, 1.2-mm-wide gaps were introduced to accommodate a brain-cutting knife and were placed every 4.8 mm by erasing the corresponding portion of the mask (Supplementary Fig. 1A, center, B), yielding 24 6-mm-thick slices from Patient 1 and 22 6-mm-thick slices from Patient 3. The superior half of the mask was then removed at the level of the maximum right-left span; this left behind a mask of the inferior skull base surface of the brain (Supplementary Fig. 1A, center). This partially masked image was then inverted to obtain the shape of the cutting box (Supplementary Fig. 1A, B, right column). Because this process can be easily copied and repeated for all slices in the 3D T1-MPRAGE using image processing algorithms such as MIPAV, the 2D design of a brain-cutting box can be generated in within 2 hours. Coronal, sagittal, and transversal views of the cutting box design from Patient 1 are shown in Supplementary Figure 1C. The surface of the cutting box (2D design) was then 3D-rendered using OsiriX v5.6 (3D surface rendering tool; http://www.osirix-viewer.com/) and saved as an .stl (Surface Tasselation Language - a 3D file format compatible for 3D design software and 3D printers).

The cutting box was printed using a 3D printer (Stratasys Dimension Elite, Fortus 360mc, Stratasys, Eden Prairie, MN) in ~100 hours. The material used to make the cutting box (cost ~$400) included a plastic polymer (Acrylonitrile Butadiene Styrene [ABS]) and a support material (Terpolymer of Methacrylic Acid, Styrene and Butylacrylate) that was dissolved in a hot water solution of NaOH in ~24 hours. Using the Fused Deposition Modeling technology, the 3D printer extruded and deposited the molten plastic polymer and the support material in layers to build the object from the bottom up. The layer resolution implemented was 0.007 inches.

Before accommodating the fixed brain within the cutting box, the inner surface of the box was smoothed using a rotating sandpaper tool (Dremel tools, Robert Bosch Tool Corporation, Mt. Prospect, IL) to prevent possible tissue damage.

Sectioning procedure

The sectioning was performed by 2 neuropathologists. After the brain was placed within the cutting box, the slices were cut consecutively from the center (Fig, 1B) toward the occipital lobe and subsequently toward the frontal lobe. Slices were removed and labeled immediately after cutting. For comparison, the second brain was cut according to the traditional procedure, without a cutting box (14); this yielded 15 ~1-cm-thick slices.

Qualitative MRI-pathology matching

The quality of the match between the gross anatomy of the slices and the coronal T1-MPRAGE data was determined visually according to cortical and ventricular profiles by agreement of 2 evaluators (Martina Absinta and Govind Nair). The match was judged to be unsatisfactory when the superior and inferior, or right and left, edges spanned several T1-MPRAGE slices, indicating a mismatch of more than 0.5 mm. Because all the 3D high-resolution MRI sequences (multiGRE and FLASH) were planned using the same slices as the T1-MPRAGE, and linked via information in the DICOM image files, the correspondence of the high-resolution MRI sequences naturally followed from the correspondence of the T1-MPRAGE to the slice surfaces. To refine the match, distortions arising from tissue fixation and MRI acquisition were corrected using 2D-registration between digitized gray scale photos of the brain slice surfaces and corresponding MRI slices. The method of registration was affine (linear, 12 degrees-of-freedom), followed by a landmark thin-plate registration in MIPAV (Supplementary Fig. 2).

Brainstem-cerebellum

To assess the performance of our approach in a more difficult situation, in Patient 1 the same procedure was repeated to make a cutting box for the brainstem-cerebellum (Fig. 1C), yielding 9 6-mm-thick transversal slices, parallel to a plane passing through the middle cerebellar peduncles. After the brainstem-cerebellum was placed within the cutting box, the slices were cut consecutively in the craniocaudal direction and the comparison with the postmortem MRI was made according to the cerebellar cortex profile and the surface of the brainstem.

Histology

Block #1 was from patient 1. The selected formalin-fixed tissue block from the left temporal lobe was immersed in 30% sucrose at 4°C for ~2 days for cryoprotection and then embedded with Optimal Cutting Temperature compound for freezing and sectioning (OCT; Tissue-Tek, Sakura Finetek Europe B.V., Alphen aan den Rijn, The Netherlands). Twenty-five 10-μm-thick frozen sections were obtained on a sliding microtome-cryostat and mounted on 1 × 3″ glass slides, 6 of which were stained with hematoxylin and eosin (H&E) and Luxol fast blue-periodic acid Schiff (LFB-PAS) and compared with the MRI from the same location. Stained sections were digitized at 20X magnification using a slide scanner (iScan Coreo, Ventana Medical Systems, Inc., Tucson, AZ). In addition, immunohistochemistry for myelin proteolipid protein (PLP) using mouse IgG2a anti-myelin/PLP (Abcam, ab118484, Cambridge, MA) primary antibody and goat anti-mouse IgG2a-Alexa Fluor 647 (Invitrogen, A21241, Carlsbad, CA) as second antibody was performed on a representative frozen section to detect demyelinated cortical lesions. Further details are described in the Supplementary Methods.

Block #2 was from Patient 2. The selected formalin-fixed tissue block from the left frontal lobe was processed routinely and embedded in paraffin; 30 10-μm-thick sections were obtained on a microtome. Six of the 30 paraffin-embedded sections were stained with H&E and LFB-PAS and compared with the MRI (Supplementary Fig. 3).

Block #3 was from Patient 3. The selected formalin-fixed tissue block from the left hippocampus was processed routinely and embedded in paraffin; and 30 10-μm-thick sections were obtained on a microtome. Four of the 30 paraffin-embedded sections were stained with H&E and Cresyl violet and compared with the MRI (Supplementary Fig. 4).

In all cases, H&E-stained sections were used for overall assessment of cellular changes, neuronal and neuroglial distributions, and cortical architecture. LFB-PAS stained sections were used for assessment of myelin and demyelination. Cresyl violet-stained sections were used for assessing neuronal distribution in the hippocampal sections.

RESULTS

Brain sectioning with and without cutting box

The postmortem brains were placed in the imaging container and immersed in Fomblin, after which air bubbles were aspirated and images acquired at 7 tesla (Fig. 1A). Two brains underwent sectioning with the cutting box (Patients 1 and 3) and one without (Patient 2) (Table). The fixed brains fit without perceptible wobble or motion within the cutting box that was designed from the postmortem T1-MPRAGE images and printed on the 3D printer (Fig. 1B, C, for forebrain and brainstem-cerebellum, respectively, of Patient 1).

After sectioning, the gross anatomy of the anterior and posterior surfaces of each slice was easily matched (~1 hour of labor for the entire brain), with the corresponding MRI slices (Fig. 2A, patient 1; Supplementary Fig. 5, Patient 3). Reformatting the MRI images in transverse plane allowed visualization of the thickness and skewness of the cutting planes. There was a marked improvement in uniformity of thickness and skewness of the slices obtained with the cutting box (Fig. 2B, Patient 1) compared to the traditional sectioning method (Fig. 2C, Patient 2). In both brains sectioned with the cutting box, 1 cut was missed, resulting in 1 less slice than planned. It should be noted that the matches between the brain slice surface and MRI were not perfect for all slices: 6 of 22 slice surfaces in Patient 1 (Fig. 2B, dotted lines) and 5 of 20 slice surfaces in Patient 3 were judged as not accurate. This was thought to be due to slight movement of the brain during the sectioning process.

Figure 2.

Comparison of sectioning performance with and without the cutting box. (A) Comparisons of the gross appearance of the brain slices (anterior surface, starting from slice 2) and the corresponding coronal multigradient-echo (GRE, 2nd echo) magnetic resonance imaging (MRI) slices of the brain of Patient 1. Areas where the match was judged less accurate are indicated by asterisks (*). (B, C) The cutting lines and their accuracy with the cutting box for Patient 1 (B) and without the cutting box for Patient 2 (C) are superimposed on an axial MRI slice. The sectioning is much less accurate with the traditional cutting method (C).

By comparison, the performance of the traditional cutting method (Patient 2) was poor, with 13 of 15 ~1-cm-thick slices (86%) judged to be non-parallel to one another and to the plane passing through the mammillary bodies (Fig. 2C). Matching of these slices to the MRI was therefore extremely difficult and time consuming, requiring ~12 hours of labor for the whole brain.

Supplementary Fig. 2 offers a closer look at the match between MRI (T1-MPRAGE on left) and the anterior and posterior surfaces of slice 13 (Patient 1). Optimal registration was achieved using 2D-coregistration between gray scale photos of the slice surfaces and corresponding slices on MRI.

Brainstem-cerebellum sectioning with the cutting box

Because of its success in improving the accuracy of pathological sectioning of the cerebral hemispheres, we considered whether the same approach might be applied to a more difficult problem: the brainstem-cerebellum (Patient 1). Two of 8 slices (25%) were judged to be insufficiently accurate (Fig. 3, indicated with *). As with the cerebral sections, fine structures, including demyelinating lesions centered on prominent veins (Fig. 3, arrows), could be appreciated on the MRI images (Fig. 3, in gray scale) and easily colocalized in the gross examination (Fig. 3, color photos). Direct comparison between the sectioning performance of brainstem-cerebellum with (Patient 1) and without cutting box (Patients 2 and 3) is not available.

Figure 3.

Brainstem-cerebellum sectioning performance with the cutting box. Matches between the gross appearance of the slices (left) and corresponding transversal multigradient echo (GRE, 2nd echo) magnetic resonance image (MRI) slice of the brainstem-cerebellum are shown for Patient 1. White arrows indicate demyelinated lesions; in several of them, the central vein is prominent on the GRE MRI sequence. Areas where the match was judged less accurate are indicated by asterisks (*).

MRI-guided histopathology

To assess the correspondence between high-resolution postmortem MRI and histopathological findings, we chose a sulcus (slice 18) in Patient 1 that appeared to harbor a cortical lesion on both in vivo and postmortem scans (Fig. 4A, B) (15–17). The lesion was better appreciated on the postmortem GRE T2* (Fig. 4B, C) than on the in vivo images, probably due to better resolution of the postmortem scan and to changes in the MRI relaxation properties related to fixation. The histological characterization and the comparison with the MRI are shown in Figure 4C–E, where white arrows have been manually drawn to indicate the lesion extent and a red line to show tissue missing from the histology image (lost during cryosectioning). Furthermore, the general shape of the sulcus, as well as the location, size, and shape of both the lesion and sulcal blood vessels, are extremely well matched between MRI and histologic analysis (Fig. 4C–E), demonstrating the accuracy of the technique. As shown with immunofluorescence staining for PLP, the normal myelin architecture of the cortex is disrupted at the level of the lesions due to the lesion that affects the cortex from the pial surface to the approximate location of cortical layers IV-V (Fig. 4E). This lesion can be classified as a type 3 subpial cortical lesion (18).

Figure 4.

Multimodal magnetic resonance image (MRI)-histology examination. (A–E) A cortical multiple sclerosis lesion was barely visible on the in vivo scans (A, boxed area) but clearly visible on the postmortem 3D multigradient echo (GRE) scan (B, white box) and (C). White arrows indicate the extent of cortical lesions on the GRE MRI image (C), and in histologic sections with H&E staining (D) and myelin proteolipid protein immunohistochemistry (E). The red line in (C) indicates where the fixed tissue was broken during slicing; this is contour is evident in the right lower portion of (D).

Similarly, the histological characterization and the comparison with the MRI were performed for hippocampal sections (slice 13) in Patient 3 (with anti-NMDA receptor encephalitis), and shown in Supplementary Figure 4. By H&E and Cresyl violet staining there were no remarkable histological findings in the hippocampi of this patient.

For comparison, a third tissue-block was sectioned in the left frontal lobe (slice 6, Patient 2) of the brain that underwent the traditional cutting procedure (14). The histological characterization and the comparison with the MRI a posteriori are shown in Supplementary Figure 3A–D. Even with the larger section (i.e. 2″ × 3″ vs. 1″ × 3″ glass slides [Fig. 3]), which in principle would improve topographical localization, the match was less accurate and more time consuming. This difficulty was compounded because the cortical architecture was partially missing due to tissue sampling and/or loss during histological processing.

DISCUSSION

Combining different expertise and multimodal approaches can contribute to advance our knowledge in understanding the pathobiological basis of different neurological diseases. Here, using an individualized cutting box for fixed-brain tissue rather than traditional sectioning approaches, we show that an accurate multimodal integration between MRI and pathology is feasible in the research setting.

Postmortem MRI acquisition of fixed brain tissue is useful because it provides a more detailed view of brain structures and identifies minute findings that may have been missed in vivo; thus, it may facilitate better MRI-neuropathological analysis for clinical-pathological correlation (8, 19). Nevertheless, postmortem scanning is challenging for several reasons. First, it needs to account for changes to the T1 and T2 relaxation time constants due to formalin fixation (20–25). Second, it is difficult, for technical reasons, to image the whole brain with multiple optimized imaging contrasts at high resolution in a single setting. Conventional studies of MRI-pathology correlation involve imaging slices of tissue after the initial cutting of the brain (1, 2 l, 3, 5, 19, 26). In such cases, higher resolution scans can be performed on individual slices in the same imaging time due to the limited coverage necessary. In our experience, several advantages prompt the preferential MRI acquisition of the whole brain over single or multiple ~1-cm-thick slices; most important among these is the preservation of anatomical landmarks, which facilitates reliable comparison with the longitudinal in vivo scans.

Our results clearly show that sectioning is superior using the cutting box (Figs. 2, 3; Supplementary Fig. 5). Matching the pathological sections and MRI scans was easy and rapid (~1 hour vs. ~10–15 hours required after standard sectioning), because the comparison was direct and it was not necessary to reformat or resample the scan. Following visual identification of matching slices, further digital co-registration between MRI and pathology can correct residual inaccuracies, if necessary, prior to quantitative analysis. In addition, the targets assessed herein – a cortical MS lesion (Patient 1) (16, 17) detected on postmortem MRI (Fig. 4) and the hippocampus (Supplementary Fig. 4) – were correctly localized after the pathological cut in the expected slice and at the expected location. These targets were subsequently analyzed using standard histological stains, confirming that a complete translational approach from in vivo to postmortem data is both feasible and valuable.

Mismatches between in vivo and postmortem MRI also arise due to distortions from subtle motion between hemispheres. For these reasons, care is required to identify the sulcus and other regions of interest correctly in the postmortem images. For histopathological correlations, additional important sources of error can be the discrepancy between the thickness of the MRI slice and the histological section, the related issue of volume averaging, and tissue shrinkage, which may be method-dependent. It has been reported that fixed tissue shrinks by 15% to 30% upon paraffin embedding (27), which would have to be accounted for when measuring the distance from the slice surface to the lesion of interest in MRI-guided histology. It should be noted that although widely variable among tissue types, such shrinkage and distortion is reportedly less with the cryosectioning technique than with other procedures, such as paraffin embedding (28).

The 3D cutting box method can be extended beyond what we report herein, for example, by changing the orientation of sections to match disease features or to reduce slice thickness below 6 mm. The design flexibility that is intrinsic to our approach is the major advantage in comparison to the non-customized slicing box for brain hemispheres that has been implemented in previous MRI-pathology correlation studies (4, 10). In this context, research studies that require maintaining the anatomical integrity of midline brain structures will benefit from the whole-brain approach.

Of note, the feasibility of this method is not limited by the strength of the MRI magnet used because the anatomical sequence used to design the cutting box (~6.5 minute 3D T1-MPRAGE) can be easily acquired on 1.5 and 3 tesla scanners. In our experience, a short training in the use of this device helps to minimize sectioning mistakes (e.g. missing, wavy, or incomplete cuts), as occurred for a few slices in Patients 1 and 3. Simultaneous sectioning of all the slices instead of sequential cuts might reduce skewness and other errors but this would require a customized cutting device (e.g. multiple wires or knives set at the interslice distance). Finally, in the implementation described here, localization and histological analysis with small tissue blocks and 10-μm-thick sections rather than whole mounts could also be achieved using a similar technique.

In conclusion, our approach embodies the concept that whole-brain postmortem MRI can facilitate and guide pathology thereby providing a benchmark for comparison before and after sectioning. The use of an individually rendered, 3D-printed cutting box for fixed brains can improve the speed, quality, and accuracy of pathological localization of small lesions identified on MRI, such as commonly occur in MS and many other brain diseases. The devices tested in this study (i.e. imaging container and cutting box) offer a clear methodological improvement and can be easily adapted for use in other brain disorders and optimized according to specific research questions.