I was lucky enough to attend the AI and Machine Learning for Computational Medicine conference organised by Big Data and Computational Oncology Lab on Tuesday 27th March 2018. This was as part of the inaugural Computational Medicine module run by Dr Matt Williams and Dr Caroline Morton (if you follow this blog those names may be familiar to you!)
There were a wide range of talks lined up, covering various computational models and Bayesian inference, argumentation-based clinical decision support systems, machine intelligence in image analysis, network analysis to study surgical innovation and more, so we were promised an intriguing session from the get go.
The conference started with a talk from Dr Rajesh Jena, a Computational Oncologist from Cambridge who blew us away with beautiful visual representations of the brain. The accompanying talk was equally engaging. He had computationally modelled glioblastoma growth using a technique called diffusion tensor imaging. This is where water diffusion is used to determine where certain highly structured tissue lies. The model used MRI imaging to detect both isotopic water diffusion, where water has unrestricted movement, and anisotropic water diffusion, where water movement is restricted by structural barriers in structured tissue. Water diffusion can also be used to class a growing glioma as low grade, which has slower growing larger cells with higher water content and therefore more diffusion occurring, or high grade tumours, where the opposite is true. He finally talked about using the Monte Carlo method to produce a heat map predicting high risk areas for tumour migration. This is useful because it allows for more targeted radiotherapy, maximizing cancerous cell destruction and minimizing healthy tissue damage.
Another very different but equally interesting computational application was ROAD2H, a healthcare resource optimization tool, using argumentation to support clinical decision-making, which was shared by Francesca Toni, Professor of Computational Logic from Imperial College London. Her team made evidence-based guidelines for prescribing certain treatments arguments, then used argumentation to recommend the best clinical decision in a given scenario. The argument strength was determined by sharing the debate in a system and mapping both responses and the argument itself to nodes. By using the number of nodes connected to the argument node, the argument strength could be determined, with more connections to the argument node indicating a stronger argument. This support system is especially useful as it allows clinicians to see reasoning behind program conclusions, allowing doctors to explain to patients why a certain course of action is recommended.
I found it really encouraging that there was a general consensus that became obvious during these talks, that coded programs used in research should be made publicly available. It gave me a great respect for the commitment of researchers to solve computational problems in medicine, and their ability to work as a community to increase problem-solving efficiency.
The day ended with an interactive panel discussion; a chance for postgraduate attendees from various fields to bring up difficulties they had encountered and troubleshoot. Although still only a medical student myself, (albeit one in danger of becoming comp-med obsessed,) it was interesting to hear about some of the challenges faced by doctors at various stages of both their computational and medical careers, and the responses given by the panel to the issues raised. A recurrent issue of Computational Medicine covering two diverse specialties meant individuals were not aware of all the useful resources available, having had no formal foundation in programming to put their computational exercise in context. It sometimes meant that researchers were stuck with a problem, when the solution was very likely to be out there. The panel had multiple solutions to this situation; some had employed a computer scientist to come in, who had a large knowledge bank in the relevant field. However, this did not always solve the problem as those available didn’t necessarily have the right subspecialty knowledge to help since computing is such a large field of study. Others suggested that organizing more frequent informal meet-ups to fill in some of these knowledge gaps might be a better solution. The audience agreed.
More than anything, the conference was a satisfyingly compact taste of how academics have been applying their programming ability in research. Just as importantly though, I left feeling a real sense of community, which is comforting in a reasonably novel and not quite yet mainstream field where I anticipate it is easy to feel isolated in what you do.
All in all, a most gratifying day out. A big thank you to everyone who made it happen!
Written by Aditi Basavakumar (4th year medical students, Imperial College London)