In the latest episode of ESR Connect Weekly, Prof. Andrea Rockall, from the Imperial College London, presented a use case that she is currently developing with her team. She also invited the audience to send suggestions and ideas to email@example.com in order to get them involved in the project.
As an introduction, Rockall showed two patient cases with suspicion of oncocytoma or renal cell carcinoma. She had the audience decide which one was which and almost 60 per cent were misled by the images. This underlined the clinical issue behind the use case, as more than 60 per cent of all renal cancers are actually diagnosed incidentally. It is hard to tell the difference between a malignant renal cell carcinoma and a benign oncocytoma. There is currently no reliable diagnostic technique that has sufficient accuracy in tests to allow a confident distinction between the two, neither is there a cross-sectional imaging technique, PET tracer, nor circulating tumor marker. Rockall underlined these facts with the latest literature on this issue. Thus, an algorithm that can differentiate between both would be extremely useful.
As there is a clearly unmet clinical need, Prof. Rockall suggested writing an algorithm to reduce the number of biopsies and surgical excisions or ablations of what may turn out to be a benign lesion. An AI tool could help to improve the diagnostic performance and prevent oncocytomas being operated upon unnecessarily, which would not only be more economical but would also be less burdensome for the patient.
As co-hosts, Prof. Rockall welcomed Wenjia Bai and Sayinthem Vivekanantham to the studio. Dr. Bai is a machine learning expert who also works at the Imperial College London and explained what type of information he ideally needs from the clinicians in order to develop algorithms, which essentially consists of a large number of very diverse data. Subsequently, Mr. Vivekanantham explained how data is lost in the clinical process. Prof. Rockall thereby exemplified the gap between the ideal requirements and the actual work and she discussed what should be considered for machine learning projects from a clinical perspective.
This week, the popular recurring feature Theo’s Technical Top Tips focused on splitting data into training-, validation- and test-sets. Theo Barefoot, with whom you may be familiar from previous episodes of Reasons to do AI with Friends, is an MRI physicist at the Royal Marsden London who presents one valuable detail per episode on what is necessary for any machine learning project.
Andrea Rockall then presented the workflow requirements for developing this machine learning tool in detail, and what information can be used on top of the imaging information when developing a database. She followed the workflow requirements with an explanation of what radiomics means and what steps are necessary for the validation process before the clinical validation. As a final point, she told the audience about what the final output of the tool will hopefully look like. Naturally, the challenges in projects like this were mentioned.
At the end of the presentation, the hosts spent an extra fifteen minutes to answer open questions that the audience posed in the live chat. Prof. Rockall will now take a brief hiatus from live shows for a week and return for the final episode of Season 2 of ESR Connect Weekly on Wednesday, February 19.
This season, Reasons to do AI with Friends is broadcasted live every Wednesday and presents use cases covering a variety of issues. The next episode, “The one where we find the connection”, will be hosted by Georg Langs and Mariana Diogo and will be broadcasted from Vienna, Austria on February 12 at 19:30 CET. Registration is free of charge.
All episodes will be available on-demand for purchase on connect.myesr.org.
Read more on Episode 1, “The one where we classify breast cancer with MRI” here.
Read more on Episode 2, “The one where we tackle fibrosis” here.
Read more on Episode 3, “The one with the sixpack” here.
Read more on Episode 5, “The one where we find the connection” here.
Want to deepen your AI knowledge? Check out the reading recommendations from our experts here.