AI News, The top 100 most cited articles in medical artificial intelligence: a ... artificial intelligence

Artificial Intelligence Distinguishes Surgical Training Levels in a Virtual Reality Spinal Task

With the shift toward competency-based curricula, surgical educational paradigms are evolving to include new methods of assessment and training.

Simulation has become important in surgical education, with many programs implementing courses involving animal models, cadavers, benchtop models, and virtual reality simulators2.

The requirement of efficiently combining multiple metrics has resulted in the need to assess systems that are capable of analyzing extensive amounts of information from multivariate data sets.

Artificial intelligence employs machine learning algorithms, giving computers the ability to identify patterns and perform tasks without explicit programming when sufficient data are provided6,7.

In surgical simulation, supervised algorithms could be trained utilizing sets of metrics labeled as senior or junior, thereby allowing them to classify new individuals’ metrics as senior or junior.

One-class learning (training algorithms to identify individuals belonging to 1 group [e.g., experts]) and multiclass learning (training algorithms to classify individuals in ≥2 groups [e.g., junior residents, senior residents, and staff surgeons]) also could be employed but would require a large number of participants in each group to adequately train the algorithms.

We addressed 3 questions in this investigation: (1) Can artificial intelligence uncover novel metrics of surgical performance that differentiate between 2 groups of different training levels?

(2) Can support vector machine algorithms be trained to recognize whether an individual executing a virtual reality hemilaminectomy is of senior or junior level?

Participants were divided into senior (postgraduate year [PGY]-4 and above) and junior (PGY-3 and below) groups because our group of surgeons considered that the simulated procedure required basic burr and suction instrument-handling skills that should be acquired by the fourth year of orthopaedic and neurosurgery training.

As demonstrated in Video 1, the virtual hemilaminectomy required participants to remove the L3 lamina with a simulated burr in their dominant hand while controlling bleeding with a simulated suction instrument in their nondominant hand (Figs.

{'href':'Single Video Player','role':'media-player-id','content-type':'play-in-place','position':'float','orientation':'portrait','label':'Video 1','caption':'This video demonstrates an individual performing a simulated L3 hemilaminectomy with the NeuroVR platform.','object-id':[{'pub-id-type':'doi','id':'10.2106/JBJS.18.01197.vid1'},{'pub-id-type':'other','content-type':'media-stream-id','id':'JBJS1801197V1'},{'pub-id-type':'other','content-type':'media-source','id':''}]} Artificial intelligence methodology was applied through a series of steps, including raw data acquisition, metric extraction, metric normalization, metric selection, machine learning algorithms, and model selection (Fig.

This step is vital to prevent the algorithm from receiving irrelevant input, thereby avoiding the training of algorithms that are too closely “fitted” to a specific data set and tend to generalize poorly to new subjects17.

This backward algorithm started with all of the metrics chosen by spine surgeons and removed them sequentially while iteratively training a machine learning algorithm and testing its accuracy using 10-fold cross-validation16.

Support vector machines are suited for small sample size and multivariate data that are necessary for global evaluation of surgical skill, thereby making them a prime candidate for virtual reality surgical simulation7,17,18.

In a manner similar to the coefficients in a linear logistic regression, these algorithms attribute a weight to each metric and make their classification on the basis of an equation that considers every metric and its respective weight.

Four other algorithms (k-nearest neighbors, linear discriminant analysis, naive Bayes, and decision tree) were also trained to assess whether the selected metrics could achieve a similar accuracy with diverse classification methods.

To analyze the performance of senior and junior participants, the ratio of the average metric score for senior and junior participants (the fold difference) was calculated for each metric.

Twenty-two senior participants (6 spine surgeons, 3 spine fellows, and 13 senior residents) and 19 junior participants (11 junior residents and 8 medical students) were recruited.

Finally, senior participants displayed slower deceleration overall, showed higher delays between 2 consecutive accelerations while removing L3, and exhibited less variance in the pitch angle of the burr when they removed L3.

Furthermore, the senior participants displayed less angle variance with the burr when removing L3 and higher delays between 2 acceleration peaks, which provides new insights on the consistency of their movements.

In addition, they will individually be guided to improve their skills through video-based and auditory feedback, which attempts to mimic current training in the operating room whereby surgeons explain what to improve and demonstrate how to do it.

We addressed the second research question by training a support vector machine algorithm with 12 metrics to classify senior and junior participants performing a virtual reality spine procedure.

This individual applied less force on the dura, spent more time using both tools simultaneously, and displayed more consistency with the burr (less variance in pitch angle and larger distance between 2 acceleration peaks) than other juniors.

Since spine training varies from 1 program to another and PGY-4 is a pivotal year in terms of surgical skill acquisition, efforts were made to understand whether the PGY-4 individuals should be included in the senior group.

Furthermore, if large numbers of spine surgeons are recruited, 1-class learning could be used to train algorithms to recognize expert performances and assess participants according to expert standards.

The significance of this study lies in the potential of combining virtual reality simulation and artificial intelligence to provide safer training and objective assessment of surgical skills, which could lead to improved patient care.

Stanford HAI 2019 Fall Conference - Owning AI: Intellectual Property for Artificial Intelligence

Ryan Abbott, Professor of Law and Health Sciences, University of Surrey School of Law; Adjunct Assistant Professor of Medicine, David Geffen School of ...

Stanford CS230: Deep Learning | Autumn 2018 | Lecture 8 - Career Advice / Reading Research Papers

Andrew Ng, Adjunct Professor & Kian Katanforoosh, Lecturer - Stanford University Andrew Ng Adjunct Professor, Computer ..

John Schulman: OpenAI and recent advances in Artificial Intelligence - #16

John Schulman is a research scientist at OpenAI. He co-leads the Reinforcement Learning group and works on agent learning in virtual game worlds (e.g., Dota) ...

Keynote: Knowledge Systems and AI

In an ideal world, there would be infinite computing resources. These resources would be free and sustainable, with no impact on the future of our planet.

Stanford HAI 2019 Fall Conference - Artificially Intelligent Associations

Artificially Intelligent Associations: AI, Civil Society, and Human Rights Lucy Bernholz, Director, Digital Civil Society Lab and Senior Research Scholar, Stanford ...

Shaping a 21st Century Workforce – Is AI Friend or Foe?

Visit: 0:30 - Introduction by Rui de Figueiredo 5:04 - Main Talk - Jennifer Granholm 51:18 - Audience Questions Jennifer Granholm, former ..

Black-Box Medicine: Legal and Ethical Issues

Black-box medicine—the use of opaque computational models to make care decisions—has the potential to shape health care by improving and aiding many ...

GraphConnect 2018 Keynote: Emil Eifrem and Hilary Mason

Watch Neo4j CEO Emil Eifrem present the 2018 NYC GraphConnect keynote, followed by a presentation on the future of artificial intelligence and machine ...

The Ethics and Governance of AI opening event, February 3, 2018

Chapter 1: 0:04 - Joi Ito Chapter 2: 1:03:27 - Jonathan Zittrain Chapter 3: 2:32:59 - Panel 1: Joi Ito moderates a panel with Pratik Shah, Karthik Dinakar, and ...

Learn Data Science Tutorial - Full Course for Beginners

Learn Data Science is this full tutorial course for absolute beginners. Data science is considered the "sexiest job of the 21st century." You'll learn the important ...