Master Assignment
Evaluation of AI-driven interpretation on microscopic images in oncology
Type: Master BIT/DSB or CS/DST; Computer Vision expertise is a plus
Period: TBD
Student: (Unassigned)
If you are interested please contact :
Introduction:
Within oncology, more and more use is made of targeted therapies that require so-called predictive diagnostics. Part of the diagnostics concerns DNA and RNA research and another when immunohistochemical assessment must take place before targeted therapy can be initiated. A good example is immunotherapy for which a PD-L1 determination is required in lung oncology. This PD-L1 determination is performed on tumor sections that are visually classified by a pathologist with a PD-L1 score of <1%, 1-49% or >50%. Unfortunately, this method is rather subjective and not optimally reproducible when the result is around the cut-off points. In addition, recent research has also shown that even the personality of the assessor influences the assessment.
With the help of algorithms that are applied to scanned images, a more reproducible result can be achieved with artificial intelligence (AI). Applications of AI with deep-learning algorithms are emerging in the context of predictive diagnostics, but have not been validated. We hope to be able to do the development and validation in Isala. PD-L1 is a marker that serves as an example here, but there are several markers that benefit from an accurate assessment by AI, such as the Ki67 determination for breast carcinomas and neuroendocrine tumors. The assignment is therefore to develop and validate PD-L1 and Ki67 for daily use in the clinic.
An additional focus of the assignment is coloring in the microscopic images. Coloring depends significantly on the lab where it was done. One lab colors more strongly than the other and this has consequences for the AI result. Tools are now available to evaluate the coloring that has been made and to give it a certain value. The question is whether it is possible to correlate that value with the AI findings and to possibly make a correction to the AI result.
The actual development of the AI algorithm is done by an external party and is not part of the assignment. The assignment is specifically about testing the relevance, ease of use and applicability for the clinic. We are looking for a committed, smart, and critical student who wants to work with this new technology that Isala has been given as the only one in the Netherlands and to critically assess what can be improved.