Machine learning and the operating room of the future

Professor Prokar Dasgupta is internationally renowned clinician-scientist, educator, polymath and editor-in-chief of the British Journal of Urology International (BJUI). As a pioneer of robotic surgery and other technological innovations in the field of urology, he spoke to us about the central role that technology will play in the operating theatre of the future — and how it might impact on surgical training. He is also the Chief Scientific Officer of Proximie.

Twenty years ago, I performed surgery like everyone did in those days — with a knife. This was open surgery, so for the prostate or the bladder, for example, that would require a cut in the lower part of the tummy which would result in quite a bit of blood loss, a relatively long stay at the hospital and a certain amount of pain.

Technology has changed that completely. I transitioned to keyhole surgery about twenty years ago and since then have been performing these kinds of major cancer operations entirely with robotic surgery, involving small keyholes through the tummy and a camera that provides enhanced, magnified 3D visuals in HD. The end result is less blood loss, shorter hospital stays and comparatively little pain — so the way I perform surgery has transformed dramatically and resulted in better surgical outcomes.

But this is not the end point; technology — and, by extension, the operating theatre — continues to evolve, because there are still a number of unmet needs.

Prior to operating on patients, we take a lot of images such as CT scans and MRI scans. These scans are used to make the diagnosis, but we should also be using them — and even better scans, such as spectral imaging — for the purpose of image-guided surgery, which is in development and improving all the time. Another exciting technology that continues to develop is the ability to 3D print an organ for transplantation, or to plan how to carry out a prostate or kidney cancer surgery, both of which create very exciting possibilities.

Machine learning and artificial intelligence are going to have a significant impact on a number of forms of surgery — particularly in terms of dealing with large data sets and labelling videos of procedures, for example, to try and determine whether certain movements of certain instruments can lead to better surgical outcomes.

In the case of prostate cancer, you might be looking at whether there are particular ways the surgeon can move which can improve the continence or urinary control of patients. Ultimately, it’s even possible that some elements of procedures will become automated, such as stitching or suturing, as these can be carried out robotically with even greater precision. However, it wouldn’t be the case that you could just press a button and a robot would carry out the entire operation — you still need a human being to make judgements that AI or a robot simply can’t; the doctor-patient relationship remains vital, despite rapid technological progress.

A remote surgical assistance platform like Proximie has a role to play alongside machine learning and AI in achieving better surgical outcomes.

Amidst the pandemic, Proximie has become a vital tool in reducing the number of people in operating rooms, while making sure surgical expertise is available where needed — but it has a much wider and longer-term function in helping surgeons of all skill levels prepare, perform and perfect their procedures.

So, for example, an inexperienced surgeon could actually prepare for a procedure by watching their chief or their trainer perform that procedure in a video from an archive, then be guided to perform that operation on the same platform, and then have an appraisal of how they performed the procedure after it has been completed.

Platforms like Proximie therefore have the ability to build up a highly confidential data bank of video footage of surgical procedures, always anonymised and secure so as not to compromise the patient in any way. Theoretically, this video data could be labelled and annotated using machine learning, but the difficulty lies in finding a standardised approach that provides actionable results. So the first step in this process has to be ensuring that there is a clinical lead, because a machine learning expert or a data scientist alone would not know how exactly to label and annotate the footage in a standardised fashion — and if you do not have the right data then you will not get the right output. Then you have to design the right models to interrogate the data, and model design is also vital.

Still, there remain a number of unanswered translational questions. Because we are at the tip of the iceberg, no one really knows what should then be done with this data. There are a huge number of potentials: surgical planning, analysing the surgical pathway, team training using machine learning — because beyond what the surgeon is doing, you can video what the entire team around the surgeon is doing and learn from that.

“But what is very exciting to me is the translation of all this information into patient outcomes.”

For example, one of the things that has emerged in the last couple of years or so is automated performance metrics; you take all this data and put it into a ‘black box’, then use machine learning algorithms to judge the performance of surgeons at different levels of expertise.

The automated performance metrics have resulted in some very interesting findings. You would think that the most experienced surgeons will have the best results, but that is not necessarily true. Some experienced surgeons clearly have outstanding results, but equally there are some naturally talented surgeons who have outstanding results without being very high volume surgeons — and it is only a machine which can see those differences and correlate them, for example, in prostate cancer surgery, with control over urinary leakage. By doing this, the metrics can thereby highlight natural talent by revealing those surgeons who have better outcomes despite not having performed a high volume of procedures.

The whole machine learning story is just starting, and we are yet to uncover the true potential of its impact on training, or how effectively and in what way it might be able to guide the instruments of less experienced surgeons.

“There is so much that remains to be seen, but this is without doubt going to be a very exciting aspect of what will be made possible by the future of surgery.”