Detailed Description
Autofluorescence lifetime images (FLIM) reveal unique characteristics of endogenous fluorescence in biological samples, but comprehensive interrogation still relies on histology images. Co-registering these images, however, is extremely challenging due to their different nature.
The lack of availability of the ground truth for the co-registration impedes the application of many registration technologies. Our research group is interested in the development of unsupervised image-to-image translation networks that significantly improves the success of the co-registration using a conventional optimisation-based regression network. Our preliminary results also indicate that the approach is flexible enough to be applied to various image formats, for example, intensity images, showing the great potential of spectral lifetime images for rapid visual discrimination of lung cancer.
Since direct correlation of histology images with intensity/lifetime images, e.g., least square, is hardly applicable, even though human intervention is introduced, our current main interest is twofold. First, to investigate methods that could limit the currently required human intervention by finding automatic correlations and, second, to study other advanced registration techniques that are able to surpass the regression. Additionally, we found out that our developed model is highly sensitive to hyperparameter tuning, which implies that performance improvement through a more thorough search can be achieved.
Key Duties and Responsibilities
More Information
https://enzj.fa.em3.oraclecloud.com/hcmUI/CandidateExperience/en/sites/CX/job/1753