AI driven analysis of placenta pathologies in histology images (Associate Prof. Christoffer Nellaker)
PROJECT TITLE
AI driven analysis of placenta pathologies in histology images.
SUPERVISORS
Associate Prof Christoffer Nellåker
DESCRIPTION OF PROJECT
Although AI for digital pathology is a growing field of research, reproductive tissues remain understudied. Within this project we are developing and adapting cutting edge machine learning and computer vision methods to understand placenta histology images that will ultimately lead to assistive tools for slide interpretation by histopathologists.
With the key motivation to create new approaches and tools for clinical pathology and functional understanding we seek models which best represent biological meaningful information efficiently. We are investigating a machine learning approach analyzing placenta tissue phenotypes from a bottom-up approach based on biological prior knowledge. By analyzing the topological arrangement and cellular make-up of tissue sections we look to capture signals of pathologies or organ wide changes in stages. We also hope to integrate these models with additional omic and multimodal datasets, as well as direct mappings to human readable summary analyses.
The project presents a true big data problem, in that each whole slide image can be multiple Gb in size, with a billion pixels, over a million cells. Each placenta can be sampled from multiple sites, and integrating signs of pathological processes across slides presents a further complication. Together with collaborators in USA, Canada, UK, Israel and Estonia we have access to whole slide images from a broad range of clinical systems and histopathological laboratories. With these datasets we have access to a representative set of histological images of most placenta pathologies.The initial ambitions for the project will be to build a tool for quantification of preeclampsia pathology phenotypes. As an extension to this tool we will look to integrate preeclampsia histological imaging phenotypes, with spatial transcriptomics and proteomic profiles.
TRAINING OPPORTUNITIES
This project is inherently inter-disciplinary, and will require someone with a hunger to learn. As part of this project you will preferably have a background in, and an interest in learning those you don’t covering: reproductive biology, placenta histopathology, machine learning, computer vision, computational biology, python-pytorch programming, medical statistics and multi-omic integration.
HOW TO APPLY
To apply, please click here.