The Society of Interventional Oncology (SIO) hosted its first artificial intelligence (AI) hackathon at this year’s annual meeting (31 January–3 February, New Orleans, USA). Out of 11 international contestants, Leigh Casadaban (Brigham and Women’s Hospital, Boston, USA) won the competition for her project entitled “Feasibility study of deep learning convolutional neural networks to predict clinical benefit of transarterial radioembolization using pretreatment CT and SPECT imaging”.
“Our study plans to look for imaging or patient/treatment factors that predict outcome after transarterial radioembolization (TARE) with Yttrium-90 (Y-90) microspheres,” Casadaban explains. “Our hypothesis is that the deep learning system using 3D treatment planning data will outperform standard dosimetry estimates in predicting progression-free survival.”
Julius Chapiro (Yale School of Medicine, New Haven, USA) from the SIO Research Committee hosted and moderated the AI hackathon, which was funded with unrestricted grants from Guerbet and Visage Imaging. This novel type of session gave the opportunity to 11 international teams to pitch their pre-selected proposals to a jury panel, composed of international interventional oncology (IO) experts and industry representatives. The jury evaluated the proposals on the criteria of healthcare impact, innovation, translatability to product, and presentation. Each five-minute pitch presented “AI-driven research project proposals with high translational value”. The proposed projects all offered AI-based solutions to a specific clinical problem in IO, and presentations were followed by one minute of questioning from the judging panel.
Speaking to Interventional News about her win, Casadaban says: “I am happy to have won the hackathon because it validates our project idea. The prize money also helps us to afford software and hardware for data processing, and for our data scientists and software engineer collaborators.”
Some of the US$7,000 prize money will go towards a new computer with imaging software to process images, Casadaban shares, as well as statistical software for data crunching. The rest will go towards the Brigham and Women’s Hospital and Massachusetts General Hospital Center for Clinical Data Sciences, which hosts data scientists, software engineers, and clinicians working to solve healthcare problems.
AI hackathon “incredibly powerful” format in a quickly evolving field
“The hackathon was a fun format,” Casadaban says, as it is “different from other formal talks, in that presenters stood directly in front of the judges, not behind a podium”. She continues: “We had five minutes to ‘pitch’ our idea before the buzzer went off. I think the format made the session fun and exciting.
“AI already influences many things around us, and is incredibly powerful. The field is evolving quickly, and an event like this raises awareness of new developments and tools, encourages brain storming, and also enables IO projects to get their feet off the ground. We have just started to scratch the surface of how AI can improve our clinical practice. The opportunities are boundless.”
“We did not expect to collect so many excellent proposals as the AI hackathon was announced on very short notice”, says Chapiro. “The number of excellent and mature proposals and the great variety of different scientific questions raised during our session clearly illustrates the untapped potential of our community and the need for funding of structured machine learning-driven research. With this format, we can mobilise the IO community to enter the era of data science”. The SIO confirmed plans to make the AI hackathon a recurring session for upcoming meetings, including during the next SIO annual meeting, which will take place in San Francisco, USA, in January 2021.
