Towards virtual histology with X-ray grating interferometry

Abstract

Breast cancer is the most common type of cancer worldwide. Diagnosing breast cancer relies on clinical examination, imaging and biopsy. A core-needle biopsy enables a morphological and biochemical characterization of the cancer and is considered the gold standard for breast cancer diagnosis. A histopathological examination uses high-resolution microscopes with outstanding contrast in the 2D plane, but the spatial resolution in the third, Z-direction, is reduced. In the present paper, we propose two high-resolution table-top systems for phase-contrast X-ray tomography of soft-tissue samples. The first system implements a classical Talbot-Lau interferometer and allows to perform ex-vivo imaging of human breast samples with a voxel size of 5.57 μm. The second system with a comparable voxel size relies on a Sigray MAAST X-ray source with structured anode. For the first time, we demonstrate the applicability of the latter to perform X-ray imaging of human breast specimens with ductal carcinoma in-situ. We assessed image quality of both setups and compared it to histology. We showed that both setups made it possible to target internal features of breast specimens with better resolution and contrast than previously achieved, demonstrating that grating-based phase-contrast X-ray CT could be a complementary tool for clinical histopathology.

Figure 1

High-resolution phase-contrast imaging (TLI setup) and corresponding histopathological assessment of a human breast tissue (sample 1). (a) a coronal slice reveals details of the internal structure and allows to discriminate between: adipose tissue (fat cells, A—dark grey in a and white in b); connecting tissue (C—light grey in a and pink in b); milk ducts (D); multiple cysts (Cy); vessel trias (red square and c) and usual ductal hyperplasia (green square and d). This is confirmed by the corresponding histological slide (b) at the same sample orientation.

Figure 2

High-resolution phase-contrast imaging (MAAST setup) and corresponding histopathological assessment of a human breast tissue (sample 2). Histology (a) and GI phase-contrast X-ray CT (b) revealed numerous structures inside the tissue: milk duct with columnar cell metaplasia (A); dilated duct or duct ectasia (B), cysts (Cy), connective tissue with fibrosis (pink areas in a and white—in b; for example: D); fat cells (white areas in (a) and dark gray—in (b); for example: E); vessels (for example: next to F). (c)—Grating interferometer allows to reconstruct images with absorption, phase-contrast and dark-field signals from one dataset. The corresponding coronal slice in absorption CT as a comparison, showing poor contrast and an obviously lower image quality. (d)—Zoomed-in red square selection in (a). (e)—zoomed-in black selection in (d). Higher magnification on (d–e) allows to see oxalate crystals inside the milk duct (O), which are also visible as bright spots on the single X-ray CT slice (f, increased contrast compared to b–c, g) and its Maximum Intensity Projection over 20 slices (g).

Figure 3

Lower part of sample 2 (with respect to the Fig. 2, b), examined with histology and GI phase-contrast X-ray CT (MAAST setup). X-ray CT (a–b) and histological (c–d) slices are not fully mutually aligned. (a)— one slice in the tomographic reconstruction of the data acquired with MAAST X-ray setup. It reveals some features, that can be also identified at the histological cross-section (c): fat (A), connecting tissue (B), vessels (C), benign apocrine cysts (E). (b)—Maximum Intensity Projection over 50 slices of the tomogram in (a) helps to notice bright white spots with different intensities that are likely calcifications. (d)—zoomed-in area at the histological slide (red square, c) highlights benign secretory calcifications inside cysts (E), as well as several duct cross-sections with low-grade DCIS.

Figure 4

Sample 3 with high-grade DCIS was examined with histology and GI phase-contrast X-ray CT (MAAST setup). (a)—Histological examination reveals numerous malignant cells inside milk ducts (A). Less affected ducts (B), fat (D) and fibrosis (pink area in the connecting tissue; for example, C) are also visible. (b)—One tomographic slice of the data acquired with MAAST X-ray setup reveals most of the features, identified at the histological cross-section. Absorbing areas inside the necrotic ducts (A) are clearly distinguishable. (c)—denoised version of the image b allows to better see the shapes of individual ducts. A non-local Means 2D filter was applied with smoothing factor 2 and auto-estimated sigma. (d)—Maximum Intensity Projection over 1000 slices of the tomogram at (b). The image reveals numerous calcification clusters, that are an indication of the severe necrosis inside milk ducts. (e)—zoomed-in yellow square from the histological slice (a) highlights necroses inside the milk duct associated to DCIS. (f)—while necrotic formations inside the duct are also nicely visible with X-ray CT, the contrast is highly dependent on the exposure time per image in a given CT scan. (g) Deep-learning Noise2Noise-based denoising allows to clearly visualize inhomogeneities inside the milk duct with three times decreased acquisition time: the image quality of denoised 10-s image is comparable with, and even superior to the quality of the original 30-s one (SNR of 20.4 vs 15.7). (h) zoomed-in red square from the histological slice (a) highlights three milk ducts with severe necrosis. Milk ducts are surrounded by areas with inflammation (X). (i)—zoomed-in area from (h) reveals small calcifications that grow on the necrotic formations (yellow circles), they are also visible as white saturated dots at the X-ray CT slice (j). (k) Maximum Intensity Projections over 20 slices help to highlight calcifications inside ducts more prominently—calcifications remain visible even when the contrast degrades while decreasing the exposure time per image in the CT scan.