Developing a convolutional neural network for the analysis of gonadal tissue

This project aims to develop a convolutional neural network (CNN) for the automated analysis of gonadal tissue slide scans. Our team has the largest single centre biobank of paediatric oncofertility tissue stored for fertility preservation via cryopreservation. During surgery, a small portion is kept for histopathology analysis.

Histopathology analysis can help predict the success of different fertility preservation procedures. Currently, a lab technician needs to analyse the histology slides manually. This is a slow process – either counting ovarian follicles and follicle type in ovarian tissue, or determining the stage of spermatogenesis and presence of sperm in testicular tissue. We want to train CNNs to automate this manual process.

Currently there is interest in developing CNNs for ovarian tissue analysis, but most of development is from adult tissue. Paediatric tissue has a much higher rate of abnormal follicles, which would allow for a more robust training of our proposed CNN. In addition, no groups are currently working on a testicular tissue CNN, as far as we are aware. We hope to develop not only the most robust ovarian tissue CNN analysis tool, but also the world's first testicular tissue CNN analysis tool.

This phase of our ongoing research project would use standard CNN training methods, which would have a very quick impact on our department and would directly lead to innovations in clinical care. MDAP collaborators will contribute through slide annotation software and machine learning development, as well through the provision of feedback, analysis and project review.

Key outcomes include:

• Ability for clinicians to quickly analyse patient slides for more personalised care. For example, knowledge of ovarian follicle density would help guide counselling around future fertility options, such as continue storing cryopreserved ovarian tissue with its associated costs.
• Easy access to the analysis of information to reduce inequities around different Australian and international centres – giving clinicians access to information that is mostly currently absent to the decision making.
• Digital analysis of tissue will allow researchers to better understand ovarian and testicular function for the development of better models of ovarian follicle and testicular dynamics.

We aiming to establish a national oncofertility biobank as well as a national oncofertility tissue slide repository. Having CNN tools would allow oncofertility centres around Australia, as well as globally, to quickly analyse tissue to make important clinical decisions for their patients.