Hyperspectral Disease Diagnosis
Peter’s landmark paper: JT Kwak et al., “Improving prediction of prostate cancer recurrence using chemical imaging”, Scientific Reports 5, Article number: 8758 (2015). PMID: 25737022.
The diagnosis of cancers at an early stage is critical for the long-term survival of patients. For solid cancers, such as lung, breast and prostate cancer, this is currently accomplished by staining tissue samples with hematoxylin and eosin (H&E) dyes followed by histopathological examination; time to results is typically days rather than hours. Furthermore, diagnoses performed in this way are quite subjective. Indeed, if four histopathologists examine a stained tissue sample, there could be four different diagnoses! Clearly, a technique that is faster, more accurate and less subjective than H&E staining is needed.
My paper of choice details a very careful collaborative study of the prediction of prostate cancer recurrence.
For at least two decades, vibrational spectroscopists have attempted to demonstrate the feasibility of using infrared spectroscopy in medical diagnosis. In the early studies, a FT-IR microspectrometer equipped with a single-element detector was used in the mapping mode, where spectra were measured sequentially, with the sample being moved in steps of a few micrometers.
Although the results showed promise, the time required to acquire enough spectra to fully classify tissue samples was too long. Furthermore, an insufficient number of samples were usually tested, so that any results were rarely statistically significant. As a result, optimism for such measurements was not justified.
Hyperspectral imaging achieved by the interface of mercury cadmium telluride array detectors to a standard, continuously scanning FT-IR spectrometer allows thousands of spectra of tissue samples to be measured in a couple of minutes with a spatial resolution of between 1 and 10 μm. The spectrum measured at each pixel can be classified by several different chemometric algorithms (sometimes known as chemical imaging). Several research groups have demonstrated the applicability of such methods in predicting different types of cancer. For example, groups led by Rohit Bhargava at the University of Illinois (USA), Max Diem at Northeastern University (USA) and Nick Stone at Exeter University (UK) have all made remarkable progress in that area.
My paper of choice details the results of a very careful collaborative study of the prediction of prostate cancer recurrence by scientists from the US Center for Interventional Oncology at the National Institutes of Health, the Department of Pathology at the University of Illinois at Chicago, and the Computer Science, Bioengineering, and Electrical and Computer Engineering, departments and Cancer Center of the University of Illinois, Urbana-Champaign. Their results significantly outperformed those found using the most commonly applied approaches – H&E staining with classification using the Kattan nomogram or the Cancer of the Prostate Risk Assessment (CAPRA-S) score.
The paper stands out because of the combination of very high-quality spectroscopy and data processing and the collaboration of scientists from different disciplines. The paper described an approach that, in the words of its abstract, “provides a histologic basis to a prediction that identifies chemical and morphological features in the tumor microenvironment that is independent of conventional clinical information, opening the door to similar advances in other solid tumors.”
In contrast to magnetic resonance spectroscopy, where the MRI technique was rapidly commercialized and adopted in hospitals worldwide within 10 years of demonstrating its feasibility in the laboratory, the uptake of vibrational spectroscopic techniques for medical diagnosis has been slow. Nonetheless, I believe that, within the next decade, the techniques described in this paper could displace current staining techniques for histopathological analysis – or, at the very least, I expect that they will be used alongside them.
Variations on a Drop by James Nichols
A Paper to Circulate by Ian Cree
Hyperspectral Disease Diagnosis by Peter Griffiths
Diagnosis: Digital by Liron Pantanowitz
Collagen and the Colon by Miguel Reyes-Múgica
Peter Griffiths is Professor of Chemistry Emeritus, University of Idaho, Owner, Griffiths Consulting LLC, Moscow, USA