What are the challenges the network will address?
Spectral pathology is an exciting development at the research level but there are still technical hurdles that need to be overcome to realise translation into the clinic. There are also a number of challenges associated with merging digital pathology with spectral pathology, and with extracting information from the spectral domain. We will have a number of initial work packages that will be the starting point for the network, but these will evolve as the network grows and the results of the pilot projects emerge.
WP1. Challenge: How to obtain vibrational data on a clinical time frame
Infrared imaging using focal plane array detectors is still relatively slow compared with optical whole slide imaging of H&E. However the development of Quantum Cascade Lasers has revolutionised IR imaging and raises the possibility of discrete frequency imaging. This has only briefly been explored and an AI solution to the problem of identifying the discrete frequencies, and what to do about baseline correction, was not a possibility. The development of coherent Raman approaches for rapid imaging will also be explored.
WP2 Challenge: How to combine digital pathology with spectral pathology
At present, even though spectral images are obtained, the data are analysed in the spectral domain, i.e. the spatial information is not used. In digital pathology, the spatial information is used to great effect. Combining both the spectral and spatial information analysis can turn two relatively weak classifiers into a much stronger classifier. This could improve the accuracy of diagnosis significantly. However, there are a number of challenges associated with this, particularly given the longer wavelength of infrared radiation compared with the visible. The simple definition of an ‘edge’ is not the same in both techniques!
WP3 Challenge: Can AI rather than an analytical solution be used to account for spectral artefacts
In the field of infrared spectral pathology there is an issue of spectral artefacts arising from scattering effects and substrate effects. The current approach of understanding the underlying physical phenomena, and applying analytical solutions, has had some limited success, but a simpler approach would be to use an AI solution.
WP4 Challenge: Can we use AI to develop stain-free digital staining
IR and Raman are stain free, but with machine learning it is possible to recreate the image of a number of stains. This has only been achieved by a few groups, and is there is no agreed method. At present the number of digital stains is limited, but will be dramatically expanded.
WP5 Challenge: Provide data management tools and educational resources
We require easy-to-use, cloud-based, data management tools and data sharing protocols. These will facilitate frictionless collaboration within the network, and thus the transformative potential of this work.