Back to Projects

AI-enabled Imaging

Unveiling the stiffening trajectories of the heart by physics informed machine learning 

Project ID: 2022_009

1st Supervisor: Pablo Lamata, King’s College London
2nd Supervisor: Andrew King, King’s College London
Clinical Supervisor: Dr. Ronak Rajani

Aim of the PhD Project:

  • To design, develop and deploy physics informed machine learning solutions that estimate the mechanical parameters of the heart. 

Project description/background:

Heart failure is a condition that will require numerous visits through the healthcare units, and thus generates a wealth of data that ends up scattered in different information systems. It is a complex condition caused by many aetiologies. The vision is that a better management of this condition is possible by focusing on the patient journey, where the system is able to ensure the right patients get the right treatment or technology at the right time, in the right place. We at King’s Health Partners (KHP) are leading this change with our Value Based Healthcare Programme, where Dr. Carr-White is the lead on the cardiovascular care pathway with a specific focus on heart failure ([Burnhope19], [Webb18a], [Webb18b]).  

An enabler technology towards this vision is the construction of the digital twin of the heart of each patient. In health care the digital twin denotes the vision of a comprehensive, virtual tool that integrates coherently and dynamically the clinical data acquired over time for an individual using mechanistic and statistical models. We at KCL are championing this vision [Corral20], with a wide experience in the personalization of mechanistic models to the clinical data (e.g. imaging) available [Lamata16]. The approach is thus to combine the stage of AI discovery (inductive reasoning) and mechanistic modelling discovery (deductive reasoning) to reach an interpretable decision support for diagnosis and prognosis in HF.  

The PhD project work will be devoted to the construction of the digital twin for HF, specifically to the development of physics informed neural network (PINN) solutions inspired by the work described in [Buoso21] and building on our experience in the area [Arthus21]. Working with the widely prevalent echocardiographic images, the goal is to track the health status of each patient along these longitudinal visits to the healthcare system. The routine echo images used in the visits to our valve clinics, led by Dr. Rajani at our Guys’ and St Thomas’ Trust, will be used to personalise the computational avatar, the digital twin of each subject. The focus will thus be down to one of the most common aetiologies of HF, an increase in afterload due to aortic stenosis. The expected outcome is the improved ability to characterise and predict the pathways into HF and thus the identification of the optimal time for valve surgeries.  

This project would suit a student with experience in data science, informatics/computer science, bioengineering or applied mathematics, and with an interest in clinical healthcare applications of AI enabled medical imaging.  

References

  1. [Arthus21] https://www.sciencedirect.com/science/article/pii/S002199912100259X  
  2. [Buoso21] https://www.sciencedirect.com/science/article/pii/S1361841521001122  
  3. [Burnhope19] https://pubmed.ncbi.nlm.nih.gov/30744444/   
  4. [Corral20] https://academic.oup.com/eurheartj/crossref-citedby/5775673  
  5. [Lamata16] https://link.springer.com/article/10.1007/s10439-015-1439-8   
  6. [Webb18a] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6128173/  
  7. [Webb18b] https://www.sciencedirect.com/science/article/abs/pii/S0167527317345011  

Back to Projects