Within pathology, digitalisation is becoming widely implemented in the field through systems called Whole Slide Image (WSI) Scanners. WSI scanners are systems made for digitising a tissue sample, traditionally studied through a light microscope, by scanning them with a high-resolution camera. The system acquires images covering small parts of the sample, referred to as tiles, to achieve a full mosaic image of the tissue sample.
Lumito is a company that specialises in Up-Converting-Nanoparticles (UCNP) that bind to tissue that exhibits a marker of interest. These nanoparticles undergo excitation by a near-infra-red laser, resulting in an emission of light in the visible spectra, which is then captured by the camera. More specifically, the Lumito Scanner first acquires an image showing tissue morphology, then excites it with a laser to acquire an image of the marker information. By overlaying these two images, you could showcase both morphological details and crucial tissue marker information.
The time it takes to scan a tissue sample is proportional to the scanned area. Typically, the Region of Interest (ROI), i.e., the area to be scanned, is the same as the area of the sample slide that contains tissue. The ROI is determined on an overview image, which is a low-resolution image of the sample. It can either be selected manually or preferably be calculated automatically to reduce operator time and speed up the scanner throughput.
Depending on the scanner and the number of slides, manual ROI-outlining can reach as much as a 0.5 full-time equivalent (FTE), corresponding to half-time work for one person. Automatic ROI detection will create a more cost-efficient and faster workflow. A fully automated workflow also introduces a trade-off between scan time, operator time and the number of slides needing rescanning. The task is to develop an algorithm with sufficiently high accuracy where the frequency of manual correction (ROI-correction or re-scanning) and potential prolonged scan time is low enough to justify not using manual ROI outlining from the start.
- Investigate what methods are used in the field and what accuracy they typically provide. Furthermore, investigate the accuracy threshold for rescanning. What is acceptable?
- Develop and implement a learning-based method for determining ROI from an overview image of the tissue.
- Compare said learning-based method to more naïve methods like simple thresholding and other more traditional image analysis methods.
- Compare the performance of said learning-based method to the performance of a pre-trained neural network model.
- Analyse and investigate the importance of interpretability in the developed method.
- Zarella et al. (2018) – Practical Guide to Whole Slide Imaging: A White Paper From the Digital Pathology Association
- Zarella et al. (2022) – High throughput whole‐slide scanning to enable large‐scale data repository building.
- Amin et al. (2008) – Automated whole slide imaging
Who are you:
As an applicant for this master thesis project, you should be driven, creative and not afraid to take initiatives. You should be someone partaking in a master level education with experience in programming, mathematics, image analysis and machine learning. An interest in the field also adds value to your application.
To apply, please submit your CV, cover letter, and transcripts of records through our career page on our website.
To know more about the project, please contact:
Adam Belfrage, email@example.com