Developing a methodology for three-dimensional correlation of PET-CT images and whole-mount histopathology in non-small-cell lung cancer

Abstract

Background: Understanding the three-dimensional (3D) volumetric relationship between imaging and functional or histopathologic heterogeneity of tumours is a key concept in the development of image-guided radiotherapy. Our aim was to develop a methodologic framework to enable the reconstruction of resected lung specimens containing non-small-cell lung cancer (NSCLC), to register the result in 3D with diagnostic imaging, and to import the reconstruction into a radiation treatment planning system.

Methods and results: We recruited 12 patients for an investigation of radiology-pathology correlation (RPC) in nsclc. Before resection, imaging by positron emission tomography (PET) or computed tomography (CT) was obtained. Resected specimens were formalin-fixed for 1-24 hours before sectioning at 3-mm to 10-mm intervals. To try to retain the original shape, we embedded the specimens in agar before sectioning. Consecutive sections were laid out for photography and manually adjusted to maintain shape. Following embedding, the tissue blocks underwent whole-mount sectioning (4-mum sections) and staining with hematoxylin and eosin. Large histopathology slides were used to whole-mount entire sections for digitization. The correct sequence was maintained to assist in subsequent reconstruction. Using Photoshop (Adobe Systems Incorporated, San Jose, CA, U.S.A.), contours were placed on the photographic images to represent the external borders of the section and the extent of macroscopic disease. Sections were stacked in sequence and manually oriented in Photoshop. The macroscopic tumour contours were then transferred to MATLAB (The Mathworks, Natick, MA, U.S.A.) and stacked, producing 3D surface renderings of the resected specimen and embedded gross tumour. To evaluate the microscopic extent of disease, customized "tile-based" and commercial confocal panoramic laser scanning (TISSUEscope: Biomedical Photometrics, Waterloo, ON) systems were used to generate digital images of whole-mount histopathology sections. Using the digital whole-mount images and imaging software, we contoured the gross and microscopic extent of disease. Two methods of registering pathology and imaging were used. First, selected pet and ct images were transferred into Photoshop, where they were contoured, stacked, and reconstructed. After importing the pathology and the imaging contours to MATLAB, the contours were reconstructed, manually rotated, and rigidly registered. In the second method, MATLAB tumour renderings were exported to a software platform for manual registration with the original pet and ct images in multiple planes. Data from this software platform were then exported to the Pinnacle radiation treatment planning system in DICOM (Digital Imaging and Communications in Medicine) format.

Conclusions: There is no one definitive method for 3D volumetric RPC in nsclc. An innovative approach to the 3D reconstruction of resected nsclc specimens incorporates agar embedding of the specimen and whole-mount digital histopathology. The reconstructions can be rigidly and manually registered to imaging modalities such as ct and pet and exported to a radiation treatment planning system.

Keywords: Radiology pathology correlation; non-small-cell lung cancer; positron emission tomography.

Figures

FIGURE 1

Complete pathologic response following induction chemotherapy and chemoradiation. A fibrotic rim surrounds a necrotic core. The preoperative 18F fluorodeoxyglucose pet scan demonstrates corresponding areas of rim enhancement and a photopenic centre (arrows).

FIGURE 2

The utility of embedding specimens in agar after initial formalin fixation was studied. In selected cases, it helped to maintain the conformation of sections.

FIGURE 3

The resected specimen is sectioned after insufflation with 10% formalin and is photographed with an overlying grid (pale tumour is seen on the right of each section).

FIGURE 4

After additional processing, whole-mount sections 4 μm in thickness are cut and stained with hematoxylin and eosin.

FIGURE 5

The gross section is photographed, and the lung parenchyma, visible tumour, and any intrinsic fiducial landmarks are contoured using Photoshop (Adobe Systems Incorporated, San Jose, CA, U.S.A.).

FIGURE 6

The contours from digital photographs are transferred into MATLAB (The Mathworks, Natick, MA, U.S.A.) and expanded to generate a three-dimensional rendering of the resected specimen. A contiguous outer surface relies on conformation of the sections being retained between sectioning and laying out the specimen. No extrapolation was used between sections. Green = reconstructed tumour; red and pink = surrounding lung parenchyma.

FIGURE 7

Digitised histopathology is generated from the whole-mount sections. The resolution can be varied. We determined that 2 μm was adequate for identifying microscopic disease.

FIGURE 8

Rigid (manual) registration in three dimensions of the maximum projections on computed tomography (ct, orange) and gross tumour as identified on digital photograph (blue).

FIGURE 9

Manual registration in three planes between tumour (blue contour) reconstructed using digital photographs [Photoshop (Adobe Systems Incorporated, San Jose, CA, U.S.A.) and MATLAB (The Mathworks, Natick, MA, U.S.A.)] and a fused pet–ct image.

FIGURE 10

Manual superposition of whole-mount histopathology and corresponding digital photograph. Red and green contours on the photograph represent specimen edge and naked-eye gross tumour respectively. This superposition illustrates that some mismatch remains, which could be the result of changes in conformation and specimen dimension during sectioning and processing.