Mariana Schliebs (née Patiño): Looking back on the MSc in Applied Digital Health programme

Taught terms – intense and rewarding weeks 

The Master’s programme structure made it very easy to fully immerse in the wide topic of Digital Health. Terms were organised with an intense week of full-time lectures followed by a week dedicated to the course evaluation, usually a ~3000 -word essay. This meant that over the course of two weeks I was able to concentrate on one aspect of digital health at a time, acquiring the fundamental learning during the teaching week and diving deeper into my interests during my research in the assignment week. This rhythm, paired with interesting talks taking place across Oxford, social events characteristic of the University (formal dinners, balls) meant that term time was very busy, yet also very rewarding. I took part in the Oxford University Lawn Tennis Club and attended training hours every week, so my schedule was quite intense!  However at the end of every term, I would look back at my very packed calendar and felt proud and grateful for all the memories, insights and ideas from those weeks. 

My favourite modules were AI for Efficient Healthcare Systems in Michaelmas taught by Professor David Clifton and Dr Lei Clifton  and User Focused Design and the Lifecycle of Digital Health Innovation taught by Dr Nicola Newhouse and Dr Max Van Kleek.  In the latter, we had a unique hands-on experience developing fitness solutions catered to undeserved populational groups. We were allowed to pick a 'niche' segment to study. I chose to develop the concept and user interface of a gentle workout app for postpartum women over 40 (combining postpartum recovery with bone strengthening, which is especially relevant in perimenopause). Because of the short period of time, we didn't manage to perform interviews, but I was lucky to do it with my oldest sister who was expecting her first child at the age of 43. Additionally, I got to analyse many threads in Reddit to get a sense of the sentiment - that was fun and interesting!

Mariana (centre) pictured with fellow students in class, Maryam (left) and Bill (right)

Although the teaching was incredibly rewarding, a major highlight of the programme was the people I got to share the experience with. I am leaving not only with new friends from my college and programme, but also with new inspiring mentors. We spent a lot of time together during the intense terms and we all got along very well. The small size of our cohort made it easy to build relationships with each other as well as our modules leads and programme directors. I know these friendships and bonds that will last a long time. 

Mariana and the MSc in Applied Digital Health 2024-25 cohort at their end of course dissertation presentations event and meal celebration

Research and my dissertation – creating an LLM that can interpret heart signals 

For my dissertation, I wanted to further develop my ability to take on something hard, unfamiliar and technical, and get better at it by doing it. While methods and statistics have always been my comfort zone, I hadn’t coded in over three years and my previous research leaned more towards classical statistical modelling than machine learning. My dissertation project was a perfect opportunity to push those boundaries: to take a large, complex challenge, break it into manageable steps and build a solution from the ground up. It was a true privilege to have the time and support network as I did during the thesis writing process of the MSc dissertation. I am grateful to my supervisor Dr Xiao Gu for his support throughout the process.

I set out to test whether a general large language model (LLM) could interpret Electrocardiogramme  (ECG) printouts without relying on raw signal data or a full retraining. The result was a prototype system that works like a chatbot: when given an ECG image and a question, it searches a library of thousands of cases to find the most similar examples and combines them with concise textbook notes. This approach not only improved the accuracy of ECG interpretations against the generalist model baseline in a 'low-cost' way but also make the output more transparent for the user, as the model shows the sources it draws from and the user can check the reasoning behind the output.

The system is still far from production-ready and would need to account for many additional factors before being deployed in practice, but given that it works based on pictures of the ECG printout, something standard of ECG devices across types and brands, it illustrates how such tools could eventually make ECG expertise more accessible in low-resource settings independent of the technology used for the ECG measure. 

I am proud of having completed this ambitious project, developed something from scratch and contributed, even if on a small-scale to the field of cardiovascular health, which remains one of humanity’s biggest health challenges.