Seven years ago, I started working on my thesis.
On October 25th, 2022, I'll be publicly defending it. It took a bit of time, but I'm happy with the result!
The text is available from here: [ Dissertation]
Here are the main topics covered in the thesis
- A history of deep learning for image analysis, and a presentation of the main components of deep learning pipelines.
- A history of computer vision in digital pathology, and the state-of-the-art of digital pathology image analysis before deep learning.
- The state-of-the-art of deep learning in digital pathology and its applications: mitosis detection, tumour classification/scoring, etc...
- A study of evaluation metrics and processes: which metrics are used in digital pathology challenges, what are their behaviours and limitations, how can we choose the right metric for the right task?
- A study of imperfect annotations: what is their impact on training deep networks, what learning strategies can we use when dealing with imperfections, what is the impact of the imperfections on evaluation metrics?
- An application of these concepts to the task of artefact detection and segmentation.
- A study of the impact of interobserver variability on the evaluation processes, and how to better include this variability into our evaluations.
- A description of some problems with quality control in digital pathology challenges, and recommendations on replicability and transparency for future challenges.
The slideshow (and annotated slides) of the presentation are also available: