Case Study: Chemical Imaging

Home > Product > Image classification of cell types in unstained tissue

Image classification of cell types in unstained tissue

Situation / Background

The Spero® laser-based infrared microscope was used to produce high definition chemical images of unstained thin colon tissue sections prepared with standard histological protocols. Routine chemometric analysis techniques were applied to the data in order to identify and segment tissue types.

Solution (products + services that were implemented)

Spectral image data cubes of thin (10 μm) excised colorectal tissue sections were mounted on infrared glass slides and acquired using a first-generation Spero microscope having a spectral range of 900–1800 cm−1 (5500–1100 nm), a pixel resolution of 1.3 μm and diffraction-limited resolvability set by the 0.7 NA, and a single-tile field of view (FOV) of 650 μm × 650 μm. Image cubes covering the full-band (900-1800 cm-1 at 4 cm-1 steps) as well as cubes targeting 10 sparsely selected frequencies within the band were recorded and analyzed using ImageLab® multimodal chemometrics software (Epina Software, Vienna, Austria).

To aid chemometrics data reduction, the raw image cubes were pre-processed with the following steps:

  • Noise reduction using the well-known maximum noise fraction transform
  • Rejection of pixels having an amide I peak value (1656 cm-1) of less than 0.05 AU as a spectral quality test
  • Full-band spectra converted to second derivatives to remove baselines
  • 9-point smoothing using Savitzky-Golay algorithm
  • Spectral normalization of all spectral vectors to achieve zero mean and unit standard deviation to reduce the influence of intensity changes caused by differences in cellular density and tissue thickness

Following these pre-processing steps, full band (900–1800 cm-1) second derivative datasets were analyzed using an unsupervised k-means clustering method, a non-hierarchical iterative method that obtains “hard” class membership for each spectrum. False color images for 6 classes were then constructed, whereby each pixel in the total imaged field is assigned a class membership having a corresponding color. This false coloring based on the spectral response at that pixel gives the observer a quick and easy way to “see” the biochemical differences in the sample. In this case, the main differences correspond to basic cell and tissue structure types.

Results

From the k-means clustered images, the different structures of the colon tissue are readily observable, including the colonic crypts, submucosa, and lamina propria. Within the colonic crypts, spectral signatures include combinations of mucin glycosylation bands at 1044, 1076, 1120, and 1374 cm-1 along with a strong lipid ester band at 1740 cm-1. The submucosa, in contrast, is identified by a number of strong bands that can be directly attributed to the structural protein collagen having bands at 1204, 1236, 1280, 1336 (amide III), and 1452 cm-1 respectively. Spectroscopic differences between the lamina propria and adenocarcinoma are far more subtle and are located at nucleic acid-related vibrations at ca. 964, 1062, 1090, and 1236 cm-1 respectively. The 10-frequency sparse image cube datasets were used to create high-speed qualitative chemical maps of protein, collagen , and mucin-rich regions, which can be used for guiding more detailed analyses.

Related Products


Interested in our capabilities? Let’s talk.

Want to learn more about how our laser technology can elevate your project? Tell us about it. We’ll be in touch right away.

  • This field is for validation purposes and should be left unchanged.