A network clustering approach to endometriosis diagnosis

Endometriosis is a multi-modal condition affective 1 in 10 women of reproductive age, currently estimated to be around 180 million women worldwide. Endometriosis is characterised predominantly by chronic pelvic pain, infertility and dyspareunia but can also remain asymptomatic in some women. Diagnosis can only be confirmed by laparoscopy which involves invasive surgery with associated risks. Additionally, a number of other conditions can present in a similar way which means that diagnosis can often be delayed and is often between 8-10 years from onset of symptoms. The disease is categorised into 4 stages at the time of laparoscopy and there is often a lack of correlation between the symptoms and the stage of disease. There is currently no agreed biomarkers for endometriosis.
The goal of the project is to develop a method that uses patient characteristics to predict disease stage prior to laparoscopic surgery. This method will build on the state of the art in statistical and network clustering techniques to classify patients into appropriate groups based on their symptoms and biometric profiles. Techniques will involve defining a collection of patient similarity networks, each encoding the level of similarity between each pair of individuals with respect to one relevant biometric or symptomatic feature. An optimising algorithm will then be applied to classify the patients into distinct groups so that those who are similar with respect to each characteristics included in the model are assigned to the same group and those that differ are not.
The two broad aims of this project are:-
1) To identify an appropriate collection of patient attributes to consider for clustering.
2) To develop efficient algorithms to carry out the clustering process. These must be scalable for use on large datasets, be able to deal with the multi-modal nature of the data and aim to exploit the structural properties of the data to enable computation efficiency.