Finding Cancer with Hyperspectral Imaging

Posted on By Outreach
Categories: Research Update

Paper title: A clinically translatable hyperspectral endoscopy (HySE) system for imaging the gastrointestinal tract

Authors: Jonghee Yoon, James Joseph, Dale J. Waterhouse, A. Siri Luthman, George S. D. Gordon, Massimiliano di Pietro, Wladyslaw Januszewicz, Rebecca C. Fitzgerald and Sarah E. Bohndiek

Blog Author: Jonghee Yoon

Oesophageal adenocarcinoma (OAC) is an aggressive cancer with a poor prognosis that has a 5-year-survival rate of less than 20%. If the early stages of OAC or preceding dysplasia could be sensitively detected, outcomes would be markedly improved due to the availability of non-invasive endoscopic intervention. Current endoscopic methods for OAC diagnosis include white-light, autofluorescence, and narrow-band imaging, but dysplastic lesions are difficult to identify due to poor contrast. Hyperspectral approaches, measuring tissue with many colours, have shown promise for the improving the contrast of dysplastic lesions, but currently require time-consuming sequential imaging and complex optical setups.

In order to perform spectral imaging of gastrointestinal tract in real time, we exploited a hyperspectral imaging (HSI) technique that measures both spatial and spectral (colour) information at high resolution, which is sensitive to structural as well as biochemical properties of tissue. Our hypothesis is that HSI will enable the early diagnosis of OAC by detecting distinct spectral features of dysplastic tissue, which is not achievable by conventional colour imaging methods. 

Among many HSI techniques, a line-scanning HSI method was employed, which provides a hyperspectral image with high spatial and spectral resolutions. Moreover, it enables flexible adjustments of spectral range and bandwidth of the HSI system. However, controlled imaging conditions are required to allow a wide-area hyperspectral image to be reconstructed from the line-scanning spectral images. This is challenging due to the uncontrollable movements of the endoscope under clinical conditions (Figure 1a). 

To overcome aforementioned issues, we combined the line-scanning hyperspectral system with a CMOS camera that records wide-field images for co-registration of the hyperspectral data. Co-registration of wide-field images was performed by employing a computer vision technique that extracts spatial features in each wide-field image and calculates geometric transformation matrices (GMs) by comparing features among wide-field images. Then a single panoramic image was created by using the estimated GMs (Figure 1b), which provides the information required for accurate hyperspectral image reconstruction. Therefore, the HySE system enables free-hand HSI in the oesophagus, which enables the translation of the proposed method to clinical applications.

Figure 1. Solution of uncontrolled imaging conditions under clinical environments. (a) Three types of image deformation occur during flexible endoscopic imaging: translation, rotation and magnification. (b) To account for these deformations in reconstructing a wide-area hyperspectral image from the line-scan hyperspectral data, geometric transformation matrices (GM1…n) are estimated using co-registration of the independently acquired wide-field images w1…n. Scale arrows are 60 pixels.

The HySE was found to have high spatial (up to 120 mm) and spectral (up to 0.46 nm) resolutions and give excellent colour fidelity. Encouragingly, images extracted from HySE were found to discriminate ex vivo normal and OAC tissue obtained from patients based on their spectral properties (Figure 2). We also demonstrated that HySE enables real-time (over 20 fps) hyperspectral imaging under clinical mimicking conditions using a pig oesophagus model. Our next step is to apply HySE to patients to identify abnormal features of early dysplastic lesions in the oesophagus in vivo.

Figure 2. Hyperspectral imaging of ex vivo human tissue from patients shows distinct spectral profile depending on tissue types. (a) Four representative synthesized RGB images of each sample. Dashed line indicates boundary of gastric mucosa, epithelium, submucosa, Barrett’s oesophagus, and cancer, respectively. (b-g) Spectra of the identified tissue types shown in a. Solid lines and shaded areas in b-g indicate mean value and standard deviation of the absorbance profile, respectively.  (h) Spectral angle mapper analysis using average spectral profile of oesophageal cancer as a reference signal. *** indicates p-value < 0.001. Statistical analysis was performed using a one-way ANOVA with post-hoc tests. All tissue was measured using the external illumination method. Scale bars = 1 mm.

Read more in the original journal article.