NASS 2024 RECAP
Artificial Intelligence
F. Todd Wetzel, MD
SpineLine Editor in Chief Bassett Healthcare Network, Bassett Medical Center Cooperstown, NY
At the recent annual meeting, there were two excellent sessions on artificial intelligence (AI). Given the impact this technology is having on medicine in particular and society in general, SpineLine’s editorial board felt that a brief summary of these sessions was in order.
On Tuesday, October 8, 2024, two computer scientists, John J. Hopfield and Geoffrey E. Hinton, were awarded the Nobel Prize in Physics. On October 9, the New York Times referred to this research as “Pioneering.” The Times also noted, “The award is an acknowledgment of AI’s growing significance and with the way people live and work with their ability to make sense of vast amounts of data, artificial neural networks already have a major role in scientific research…” (NYT, Oct 9, 2024, pA9).
Dr. Mark Baldwin, Assistant Professor of Computer Science at California State University Los Angeles, and a leading authority on AI, notes:
“Artificial Intelligence has rapidly emerged as a transformative force, reshaping industries, academia, and daily life through automation, predictive insights, and adaptive computational systems. The impact of AI extends across disciplines—from enhancing efficiency in professional fields like healthcare and finance to reshaping methodologies in scholarly research. AI-driven tools are becoming indispensable for solving complex problems at scale, empowering creativity, and enhancing learning. In academic environments, AI tools have found a footing as personal learning systems capable of augmenting traditional tutoring pathways through personalized and accessible experiences tailored to individual needs.
From a distance, the future of AI appears full of promise. Increased productivity, reduced human error, and expanded innovation are clear benefits that are hard to overlook. However, as we consider when and how to integrate AI into daily practices, we must also recognize its limitations. The adage 'Garbage In, Garbage Out' reminds us that flawed input inevitably leads to flawed output. In AI, the vast datasets driving decisions are largely drawn from digital representations of human knowledge. In other words, if we feed AI our biases and shortcomings, it will inevitably reflect and perpetuate them. In fields like medicine, flawed data can perpetuate bias, leading to inequitable treatment recommendations. Similarly, in education, AI-driven systems must be carefully designed to avoid reinforcing existing disparities among learners.
Encouragingly, efforts are underway to enhance transparency and accountability in AI. One promising development is the field of explainable AI (XAI), which seeks to clarify how AI systems make decisions, bridging the gap between complex algorithms and human understanding. XAI helps build trust, uncover biases, and ensure that AI outcomes align with human values. This transparency is critical in fostering fairness and inclusion, particularly in research and education. Through a commitment to accountability, openness, and clarity, we can unlock AI’s potential to not only improve individual outcomes but also create systems that uplift everyone.”
As such, it appears that the significance of these innovations would be difficult to overestimate.
On Wednesday, September 25, the first AI-related session convened, Symposium: Section on Radiology: Artificial Intelligence and Spine Surgery: Robotics, Augmented Reality, Navigation, Predictive Analytics, and other AI-Based Concepts. In the introduction to AI-based technologies, Dr. Joseph Gjolaj reviewed the definition of AI and explored areas of potential impact for AI-based technologies. Applications specific to spine care included predictive analysis, including risk stratification and treatment guidance, surgical robotics, patient specific implants, and patient monitoring. Dr. Andrew Alvarez discussed AI-based robotics, machine learning, and deep learning. One study investigating deep learning for posterior instrumentation identified 15 screws (6% of the cohort) that required revision due to violations of the proximal facet joints.
Interestingly, Dr. Alvarez noted that deep learning can work with what it is trained to do and has the potential to learn “bad habits.” This was a familiar theme throughout both symposia, with multiple speakers emphasizing the fact that they, as a surgical assistant, and despite sophisticated technology, cannot substitute for human judgment and wisdom.
In the next presentation, Dr. Andrew Mo noted that new technologies, such as navigation, robotic assisted, augmented reality, and customized implants are not always safer than traditional techniques or implants. Certainly, however, there are recognized advantages including lower radiation exposure, greater precision and accuracy, and reduced blood loss. In one study, he noted that screws positioned with a freehand technique tend to perforate the cortex medially, whereas CT navigation-placed screws more frequently perforated laterally. In one meta-analysis, looking at 37,337 screws, 91.3% were identified as accurately placed; 95.2% of the navigation assistance screws were judged to be placed accurately, as opposed to freehand accuracy of 90.3%.
Another study comparing 130 robotically inserted screws to 140 freehand inserted screws found that there was a significant reduction in the use of radiation in the robotic-guided group (62.5%). Several disadvantages were also discussed, including cost, learning curve, and the fact that many younger robotically trained surgeons may not be facile with open techniques.
Dr. Frank Mota reviewed wearables in spine surgery. Benefits of wearables include real-time data collection, data use from prior procedures, and no requirements for patient input. He did note that the data from wearables cannot yet be used alone to quantify risks or assess outcomes, but in combination with patient reported outcome measures and imaging data, the utility of wearables may be significant.
In his presentation on predictive analytics, defined as the use of data to predict future trends and events using historical data and statistical modeling, Dr. Alexander Butler began by reviewing the psychology of human decision-making. He then went on to discuss AI-based clustering of patients by a variety of factors to predict outcomes and minimize the risk of failure. This essentially resulted in process improvement with respect to patient selection and intraoperative performance. With better benchmarking and more analysis–more data are clearly needed–noise and bias can be reduced, resource utilization improved and clinically relevant predictive models developed.
The second session, Elevating Spine Surgery with AI, on Friday, September 27, was moderated by Dr. David Panias. In the first talk, Dr. Carlos Bagley, addressing AI applications for management of spinal conditions, reviewed imaging applications, emphasizing the increase in papers on AI in the literature. He then gave a broad overview of the entire landscape of AI including universal translation, robotic personal assistants, and cognitive cybersecurity. Nonoperative spine and pain management addressing gait analysis and classification of low back pain using functional models were reviewed, followed by surgical issues. Application of machine learning models to predict intra or postoperative complications were reviewed as well.
In a presentation aptly entitled “The Machines Cannot Do It All For you,” Dr. Jamie Wilson addressed critical limitations including inherent errors in machine learning algorithms and pointed out algorithm “overcall” in imaging of degenerative changes and ML models which are not robust enough to be used as risk calculators for clinical decision making. Dr. Wilson concluded, particularly in lieu of the exponential growth of AI in spine surgery, that most ML studies contain a degree of error.
Following this rather sobering presentation, Dr. Lauren Barber addressed generative AI. Generative AI is defined as the intersection of deep learning, natural language processing, and computer vision that generates content. Examples include ChatGPT, art, and media fakes. The need for clearly defined and entered prompts, followed by inference, and conclusions was emphasized.
In reviewing his experience on the use of AI for the treatment of adult spinal deformity (ASD), Dr. Neel Anand reviewed how he was enabled to approach complex cases in a minimally invasive manner and correct deformity without osteotomies. With some truly remarkable visual aids he demonstrated how adequate spine shape can be simulated based on patient morphotype. Despite impressive postoperative X-rays, Dr. Anand concluded that while AI is a tremendous aid it will never replace human wisdom.
Subsequently, Dr. Richard Skolasky addressed a somewhat broader theme. After reviewing definitions, use in surgery, imaging, recommendations, and postoperative rehabilitation, he addressed ethical considerations regarding AI. These included data privacy concerns, surgeon autonomy, and trust in systems. He concluded that AI should be viewed as a surgical assistant and that ethical and regulatory hurdles will need to develop as the technology advances.
Finally, Dr. Joseph Schwab discussed wearables in “AI infused technology: All becomes new again.” He reviewed wearable EMGs and motion tracking, electrical impedance technology and acoustic signal technology. Wearable EMGs may be useful, for example, for tracking muscle recovery after injury. Electrical acoustic signal technology signals are sent rather than received. For example, a signal may be sent to a muscle, and then recorded as it propagates through the tissue. This technology enables the physician to determine the quality of the tissue and the state of the tissue.
Large Language Models (LLMs) in orthopedics and spine surgery was discussed by Dr. Romil Shah. He looked at the utility of LLMs and patient education noting radiology reports, responses to patient questions, progress notes, and doctor-patient dialogues, which can be summarized by LLM's equivalent to physician summation in 45% and superior 36% of the time. LLM’s are also promising in terms of clinical decision making (81% accuracy answering guidelines based questions), and revenue cycle automation.
Overall, these two outstanding symposia reflected the state-of-the-art of this expanding area in superb presentations, tempered by a cautious enthusiasm. I hope everyone who attended both symposia learned is much on them and enjoyed it is much as I did.