Sci-fi is becoming reality for medicine as technology rapidly evolves. Artificial intelligence (AI) is a key factor impacting everything, including diagnosis, prognosis and management of disease. Neurosurgery, like all of medicine, will see significant transformation as a result. While AI within medical practice is not new, greater availability of big data, increasing computational power and advanced machine learning algorithms will fuel the integration of AI into future computer-based clinical decision-making paradigms. These changes will no doubt lead to several questions, from biomedical research to patient outcomes, from practical aspects to ethical concerns. Where will physicians, and more specifically neurosurgeons, fit within this new paradigm?
What has Changed
AI-based techniques, including machine learning and artificial neural networks (ANNs) have surpassed traditional statistical predictive modeling, such as multivariate logistic regression. Older AI systems utilizing an expert-based or rules-based approach require manual updating and resolving conflicts between old and new rules. Now, machine learning applies computer algorithms to data to discover patterns of interest, allowing machines to find non-obvious associations and become smarter. An ANN, a group of algorithms used for machine learning, models the data using graphs of artificial neurons, which mathematically represent the actual neuronal ability to learn. These networks can model fully integrated non-linear system relationships among many unknown variables. Subsequently, the ANN layers and weighs input data through multiple iterations to generate optimal network architecture for predicting specific outcomes.
Current Applications
ANNs within medicine currently fit into three categories:
- Diagnosis
Prognosis- Outcome prediction
Within neurosurgery, ANNs have been employed for analysis of:
- Lumbar spinal stenosis
- Disc herniation
- Brain tumors
- Epilepsy
- Trauma
- Cerebral vasospasm
Success Realized
Within the past academic year alone, the following represents a sampling of the studies that have been published:
- Outperforming traditional CT classification schemes and multivariate statistical analysis for accurately predicting six-month outcome in pediatric traumatic brain injury (TBI) patients.
- Predicting episodes of hypotension in patients with TBI offered >15 minutes warning of patient instability, potentially providing neurointensivists the opportunity to prevent dangerous arterial hypotension.
- Screening Computer Tomography (CT) brain images for acute neurologic events, such as stroke, hemorrhage and hydrocephalus, accelerating time to diagnosis from minutes to seconds.
- Heralding anterior communicating artery aneurysm rupture with a 94.8 percent prediction accuracy.
- Identifying and classifying the presence and sidedness of a perfusion deficit in patients with acute stroke.
- Improving prediction of factors leading to persistent hemodynamic depression in patients undergoing carotid angioplasty and stenting.
- Evaluating optimal DBS electrode placement for the best therapeutic effect within the subthalamic nucleus.
- Forecasting individual risk for all complications, including cardiac, wound and mortality in patients having spine surgery.
- Accurately predicting patients with chronic low back pain compared to controls analyzing spine kinematics while performing static standing tasks.
Promises and Challenges
The full impact of AI within neurosurgery cannot yet be discerned, as applications still remain in their infancy. Imagine the day when AI through ANN and other powerful technology can:
- Speed patients toward time-critical interventions
- Reduce costs through accurately predicting who will benefit from interventions
- Rapidly identify unusual cases that require urgent additional testing
- Support neurosurgeons in designing the optimal treatment intervention
In fact, the ways in which such advances would impact neurosurgery are mind-boggling! While it will never replace the humans in medical and particularly surgical care, it can enhance interventions in many ways. Clearly, machine learning will improve diagnostic accuracy, possibly supplementing much of the work of radiologists and pathologists, and will improve the ability of neurosurgeons to prognosticate and manage patients.
AI holds much promise, but also poses many challenges, from the practical incorporation of ANNs into the clinical environment to ethical considerations. Issues of concern include:
- All stakeholders in this process must consider the paradigm shift required for integrating AI into practice when determining how and to what degree AI gets integrated into the patient care workflow.
- The model is only as good as the training data, and care must be taken in addressing limitations of the model and biases that may occur.
- Over reliance on the technology is a concern, because how will clinicians catch an AI misdiagnosis? These models are complex and inappropriate reliance upon these algorithms can happen.
- There is a need for oversight into the AI development process from a socio-legal standpoint. Regulatory challenges and the burden of liability remains unclear.
- Even the best data and predictive models do not represent the individual patient and certainly do not replace the patient-physician relationship. Retaining individualized care remains important.
These models hold promise and will ultimately inform the conversation and decision-making partnership between patient and physician.
References
Azimi, P., Mohammadi, H. R., Benzel, E. C., Shahzadi, S., Azhari, S., & Montazeri, A. (2014). Artificial neural networks in neurosurgery. Journal of Neurology, Neurosurgery & Psychiatry, 86(3), 251-256.
Baumgarten, C., Haegelen, C., Zhao, Y., Sauleau, P., & Jannin, P. (2018). Data-Driven Prediction of the Therapeutic Window during Subthalamic Deep Brain Stimulation Surgery. Stereotactic and Functional Neurosurgery, 96(3), 142-150.
Donald, R., Howells, T., Piper, I., Enblad, P., Nilsson, P., Chambers, I., . . . Stell, A. (2018). Forewarning of hypotensive events using a Bayesian artificial neural network in neurocritical care. Journal of Clinical Monitoring and Computing.
Harbaugh, R. E. (2018). Editorial. Artificial neural networks for neurosurgical diagnosis, prognosis, and management. Neurosurgical Focus.
Hu, B., Kim, C., Ning, X., & Xu, X. (2018). Using a deep learning network to recognise low back pain in static standing. Ergonomics, 1-9.
Jeon, J. P., Kim, C., Oh, B., Kim, S. J., & Kim, Y. (2018). Prediction of persistent hemodynamic depression after carotid angioplasty and stenting using artificial neural network model. Clinical Neurology and Neurosurgery, 164, 127-131.
Kim, J. S., Arvind, V., Oermann, E. K., Kaji, D., Ranson, W., Ukogu, C., . . . Cho, S. K. (2018). Predicting Surgical Complications in Patients Undergoing Elective Adult Spinal Deformity Procedures Using Machine Learning. Spine Deformity, 6(6), 762-770.
Liu, J., Chen, Y., Lan, L., Lin, B., Chen, W., Wang, M., . . . Duan, Y. (2018). Prediction of rupture risk in anterior communicating artery aneurysms with a feed-forward artificial neural network. European Radiology, 28(8), 3268-3275.
Titano, J. J., Badgeley, M., Schefflein, J., Pain, M., Su, A., Cai, M., . . . Oermann, E. K. (2018). Automated deep-neural-network surveillance of cranial images for acute neurologic events. Nature Medicine, 24(9), 1337-1341.
Vargas, J., Spiotta, A., & Chatterjee, A. R. (2018). Initial Experiences with Artificial Neural Networks in Detection of CT Perfusion Deficits. World Neurosurgery.
[aans_authors]
