NSAUA 2018: Artificial Intelligence in Urology: Alexa to Terminator, “Now You Shall Cut to Heal”
In 1996 IBM’s advanced computer, “Deep Blue” beat the world chess champion by careful hand programming. Later on, the computer program AlphaZero also beat the chess champion after teaching itself in only four hours. AlphaZero’s machine-learning approach is given no human input apart from the basic rules of chess (Tabula Rasa knowledge). For the rest of the needed knowledge, it works out by playing itself over and over with self-reinforced knowledge. It hasn’t lost any matches after playing more than 100 consecutive games.
In surgery, there has been a significant paradigm shift, from the first head-light used in 1930, until the introduction of the first Da Vinchi Robotic Surgery setup joint in 1996, up to the current advanced robotic systems we use today. When assessing surgical skills, there are several factors we analyze. These include aptitudes, human factors, clinical results of end products, decisions and choices, automaticity and others.
Dr. Guru moved on to discuss the important topic of cognitive function. Data from animal studies has shown that with growing experience in storing food, birds demonstrate a growing hippocampus. This made the researchers conclude that with growing experience and learning of a specific task, there is redistribution of grey matter in the Hippocampus, with structural brain changes in response to environmental demands.
The cognitive continuum has intuition on one end of it, and analytical capability on the other spectrum. When physicians have a surgical plan before a procedure, this is influenced by clinical data, data interpretation (based on knowledge) and pattern recognition (from previous experience). This is in turn affected by other factors, including time and resources, and the environment, which allow the physicians to have situational awareness. As a result, the physicians can monitor their progress and project forward.
The decision-making strategy of surgeons is affected by several factors. These include intuitive/recognition primed (stored precompiled options), rule-based (written procedures, or rules), analytical (comparison of different options), and creative (design new option if a situation is unfamiliar). The path to true robotic surgery, according to Dr. Guru, should be the one where machines are able to recognize, process, and execute. In order to make machines understand surgical decision making, we as humans, need to understand it first.
In 2013, Dr. Guru participated in the first internal review board approved study in the operating room, using electroencephalogram (EEG) monitoring of the surgeon. In a study he published on this experiment1 no significant difference was demonstrated between expert surgeons and competent and proficient surgeons in tool based metrics except the time of the procedure, and the difference in the number of camera movements. However, significant differences were seen in cognitive metrics during advanced skills, between expert and competent surgeons.1 The metrics defined in this study included time to complete the task, the length of time either the left or the right tools were out of view, tool collision, tissue damage, distance by camera, clutch usage, distance of each hand of the robot, and task specific error.
Another important finding from this study was that the high-level engagement and mental state were the best signatures to separate all three categories of surgical expertise (beginner, competent, and expert). According to the Dreyfus model of skill acquisition, the scale begins with the beginner robotic surgeon having a narrow disjointed view, with limited understanding and poor human-machine interface. This progresses to the competent surgeon, having conceptual understanding, active decision making with optimal human-machine interface, and ends with the expert robotic surgeon, demonstrating intuition, pre-planning and sound human-machine interface.1
Two important observations are of importance. The first is that there is a clear inverse correlation between the trainee concentration and effort, and the concern and active supervision demonstrated by his mentor. The second observation was that no correlation had been demonstrated between the EEG of the mentor and the trainee reported NASA-task load index (which is a widely used, subjective, multidimensional assessment tool that rates perceived workload in order to assess a task, system, effectiveness or other aspects of performance). Both these observations were witnessed in two important tasks of robotic radical cystectomy, including pelvic lymph node dissection, and ureterovesical anastomosis of the ureter to the ileal conduit.
The next topic discussed was functional connectivity, representing the portion of brain activity which is transformed throughout the brain. Entropy is the degree of disorder and is the variability in assigning a channel to a functional community, between recordings. When expert surgeons perform various difficult surgical tasks, the channel communities are more consistent, as opposed to the situation of when a beginner surgeon performs these tasks. Moreover, the variability of channels’ communities seen in a mentor while observing a less experienced surgeon is more significant, than the variability observed in him when he himself is performing the surgery.
The next topic discussed was the architecture of brain dynamics and eye gaze that help define complex, high-risk surgical decision making (using a 120 lead EEG). In this additional study, the robotic surgeon was fitted with special eyeglasses extracting eye tracking features, and a 128 channel EEG headset while performing a robotic procedure. Data processing was done to extract all data collected from the headset and eyeglasses. This enabled to measure the communication (connectivity between channels from two separate brain areas), strength (connectivity within channels in a specific brain area), transitivity (connectivity throughout the brain), and asymmetric index (the psychological status of the surgeon). Together, these enabled the reconstruction of brain activity, and have a volumetric representation of the whole brain activity using EEG. Additionally, it enabled to measure the brain network efficiency.
The results of this study demonstrated that during an emergency event of active bleeding during surgery, we can witness higher integration with lower brain functioning autonomy. Furthermore, the surgeon’s eyes were more focused during the bleeding event, with a lower saccade rate and a higher fixation rate, which equals higher concentration and engagement.
The last topic discussed by Dr. Guru was the economy of motion and the attempt to maximize the efficiency of robotic procedures. It had been shown that honey bees avoid making choices when information is limited and the task is hard. Furthermore, they can quickly memorize the best flower location and the fastest path to it. Using this concept, Dr. Guru wanted to use a numerical analytical tool-tracking-model to actively follow surgical tools during robotic surgery, and try to machine-learn the actual movements of the robotic procedure. Using this technology, it was demonstrated that the tool trajectory is improved from non-smooth and jerky movements in novices to smooth movement shown by expert surgeons. Trying to replicate these expert movements, using different technological tools will help improve machine learning and to standardize the procedure, so that novice surgeons will be able to learn it more easily. Furthermore, it will help the robot predict the next movement and provide the support that the surgeon requires.
According to Dr. Guru, we have only just begun to scratch the surface of artificial intelligence and machine learning. One day we will be able to just sit back and watch the performance of machines in surgical sciences, but we still have a long way to go.
Presented by: Khurshid Guru, MD Chair of Urology, Roswell Park Comprehensive Cancer Center
References:
1. Guru KA et al. BJU Int. 2015
Written by: Hanan Goldberg, MD, Urologic Oncology Fellow (SUO), University of Toronto, Princess Margaret Cancer Centre, @GoldbergHanan at the 70th Northeastern Section of the American Urological Association (NSAUA) - October 11-13, 2018 - Fairmont Royal York Toronto, ON Canada