Cookies on this website

We use cookies to ensure that we give you the best experience on our website. If you click 'Accept all cookies' we'll assume that you are happy to receive all cookies and you won't see this message again. If you click 'Reject all non-essential cookies' only necessary cookies providing core functionality such as security, network management, and accessibility will be enabled. Click 'Find out more' for information on how to change your cookie settings.

Accept all cookies Reject all non-essential cookies Find out more

Artificial intelligence-based for the prognostic classification of High Grade Serous Ovarian Cancer (Professor Christopher Yau)

PROJECT TITLE

Artificial intelligence-based for the prognostic classification of High Grade Serous Ovarian Cancer

SUPERVISORS

DESCRIPTION OF PROJECT

Background & rationale:

Histology assessment by Hematoxylin & Eosin (H&E) staining of tumour tissue biopsy for the diagnosis of cancers is standard practice. However, assessment by histopathologists is time-consuming while inter-assessor variability may be a limiting factor. Several studies recently have attempted to use deep learning-based algorithms to train images for diagnosis and prognosis of cancers including High Grade Serous ovarian cancer (HGSOC)1. However, none have been clinically implemented so far, likely due to prediction models not being reproducible in multiple independent patient cohorts2.

The Oxford Classic is a 52-gene panel biomarker that can reproducibly identify poor prognostic HGSOC patients, as demonstrated in at least 13 independent cohorts of patients including the TCGA and AOCS datasets3,4. The Oxford Classic EMT-based risk stratification of HGSOC patients reliably identifies high-risk patients with significantly worse overall survival who may benefit from EMT targeting therapies and immunomodulators. Likewise, it can guide low-risk good prognosis patients towards maximal debulking surgery to improve their chances of a sustained complete remission (L. Rai, C. Yau, A. Ahmed et al (2025), Clinical Cancer Research).

In this study, by multiplex immunofluorescence (mIF) technique we identified ‘partial EMT’ (pEMT) cells in all patient tumours, however, OxC-EMT-high tumours harbor a subset of pEMT cells that seem to access and interact with stromal cells and may be responsible for chemotherapy resistance, immune evasion and metastasis. Therefore, we have a unique opportunity to develop state-of-the-art AI models to use H&E images of tumour tissues to accurately identify pEMT cells and thereby classify tumours into high or low risk of death using a common laboratory diagnostic method. We expect that this method of patient risk stratification for personalized therapy will be more affordable than molecular techniques and would require fewer infrastructure as well as trained personnel.

 

Plan of investigation:

We have generated H&E images as well as corresponding mIF images using pan-cytokeratin and Vimentin staining of tumour epithelial cells to identify pEMT cells in 22 HGSOC tumour sections (11 each of OxC-EMT high and low) from the Brescia cohort. 

We will use whole section H&E images to train and compare different deep learning-based image models to recognize cell types. Extensive evaluation of these models will be carried out. The most accurate prediction model will be chosen and tested on an external validation set of H&E images from HGSOC patients whose OxC-EMT score and molecular risk stratification is available.

 

Proposed aims and objectives:

  • Determine the accuracy, sensitivity and specificity of all tested image-based prediction models for overall survival in a training set of 22 HGSOC cases. This will determine whether deep learning-based classification models trained on H&E images can match our well-established Oxford Classic-based molecular technique of HGSOC risk stratification.
  • Determine the accuracy, sensitivity and specificity in a validation set of an additional 100 H&E images of the prediction model that performed the best in the training set. This will determine the robustness of the algorithm to risk stratify HGSOC patients.

 

Expected outputs:

  1. Identify an prediction model that can accurately and reproducibly identify high-risk poor prognosis HGSOC patients based on their diagnostic H&E-stained tumour tissue.

 

References:

  1. Breen J, Allen, K, Zucker K, et al. Artificial intelligence in ovarian cancer histopathology: a systematic review. NPJ Presicion Oncology 2023; Aug 31;7(1):83. doi: 10.1038/s41698-023-00432-6. PMID: 37653025; PMCID: PMC10471607.
  2. Piedimonte, S.; Mohamed, M.; Rosa, G., et al. Predicting Response to Treatment and Survival in Advanced Ovarian Cancer Using Machine Learning and Radiomics: A Systematic Review. Cancers 2025; 17, 336. https://doi.org/10.3390/cancers17030336.
  3. Hu Z, Artibani M, Alsaadi A, Yau C, Ahmed AA et al. The Repertoire of Serous Ovarian Cancer Non-genetic Heterogeneity Revealed by Single-Cell Sequencing of Normal Fallopian Tube Epithelial Cells. Cancer Cell. 2020; Feb 10;37(2):226-242.e7. doi: 10.1016/j.ccell.2020.01.003. PMID: 32049047.
  4. Hu Z, Cunnea P, Zhong Z et al. The Oxford Classic Links Epithelial-to-Mesenchymal Transition to Immunosuppression in Poor Prognosis Ovarian Cancers. Clinical Cancer Research. 2021 ;27 (5): 1570–1579. https://doi.org/10.1158/1078-0432.CCR-20-2782.

TRAINING OPPORTUNITIES

This project will give the student a unique opportunity to train under Professor Christopher Yau, an expert in Artificial Intelligence/Machine Learning in health care at the Big Data Institute, University of Oxford; Professor Ahmed Ashour Ahmed, director of the Ovarian Cancer Cell laboratory, Weatherall Institute of Molecular Medicine (WIMM) and a Gynecological Oncology surgeon, University of Oxford as well as Dr. Lena Rai, a senior NIHR BRC fellow with extensive experience in Cancer genomics and cellular pathology, Ovarian Cancer Cell laboratory, WIMM, University of Oxford.

The student will have the opportunity to learn state-of-the-art artificial intelligence and machine learning (AI/ML) by engaging with cutting-edge AI/ML techniques, including advanced deep learning models and computer vision applications. To this end, the student will have the opportunity to attend relevant courses offered at the University of Oxford.

Additionally, they will have the opportunity to learn laboratory based molecular genomics techniques such as RNA sequencing and digital PCR which are key molecular methods used for the measurement of the OxC-EMT.

Funding Information

This project will be fully funded by existing grants held by the supervisors

HOW TO APPLY

To apply for this research degree, please click here.