Artificial intelligence applications in neurology: seize the moment

Artificial intelligence (AI) can help to stratify patients into more precise diagnostic and therapeutic categories and enable better dynamic treatment monitoring. Prof. Kathryn Davis (Penn Medicine, PA, USA) demonstrated this in a lecture by listing numerous examples, especially from her special field of interest, epilepsy [1]. “In the years to come, we will all become much more familiar with AI.”

What are the current uses of AI in neurology? Prof. Davis had aptly decided to ask the ChatGPT chatbot, which listed the following examples: AI is used to diagnose diseases by analysing medical images, such as from X-ray, MRI, and CT, more accurately and quickly than humans. “Examples are breast cancer metastases and malignant pulmonary nodules on imaging,” added Prof. Davis. Additionally, AI-powered virtual nursing assistants can monitor and care for patients, provide medical advice, and help with medication management. Also, regarding clinical decision-making, AI algorithms can analyse patient data and medical records, assisting doctors in creating personalised treatment plans for patients. Extending on clinical applications, robot-assisted surgery is more accurate and has a lower risk of complications, thanks to algorithms that can interpret data from surgical instruments in real time. In a more fundamental research setting, AI is used to discover new drugs, identify drug targets, and design clinical trials. Also, electronic health records are analysed with AI algorithms. And then, there is remote monitoring with AI-powered wearable devices, prompting medical intervention if necessary.

Prof. Davis focused much of her talk on the use of AI in epilepsy, where it has become so important that 2023 saw the first International Conference on Artificial Intelligence in Epilepsy and Neurological Disorders (www.aiepilepsy-neuro.com). Prof. Davis extensively exemplified the use of AI in epilepsy, noting that many of these applications are “fairly easily” translated to other neurological conditions. Here are a few examples:

1. Seizures can be forecasted, as was shown in an analysis of chronic EEG data recorded with an implanted device in drug-resistant patients; seizure probability could be forecasted up to 3 days in advance [2].
2. A machine-learning method (Claritγ, Ceribell Inc.) measures the burden of seizure activity in real-time and detects status epilepticus events with high sensitivity and specificity [3].
3. Human-in-the-loop machine learning-based imaging analysis can be used to develop personalised treatment plans. This non-invasive technique detected previously hidden lesions with focal cortical dysplasia [4].
4. “AI has very good potential to change our approach to randomised controlled trials,” said Prof. Davis. An example is provided by a study in which the risk of self-reported seizure within 24 hours was forecast from e-diaries [5].
5. In the My Seizure Gauge trial, seizures were forecast and detected with wearable devices and subcutaneous EEG [6].

“In the years to come I think we will all become much more familiar with AI,” Prof. Davis remarked. The path forward for AI in neurology, she said, would include addressing the following issues:

• Bias in the dataset: “It is extremely important that data accurately represents the type of patient you intend to apply it to.”
• Errors in data collection.
• Overreliance on AI: Not all findings of AI are reliable and translatable. “Studies must be replicable and clinically meaningful.”
• The current need for regulatory guidelines on implementing AI in the clinic.
• Data privacy concerns.
• “Black box” issues: Engineers must be able to explain to clinicians how algorithms work; similarly, clinicians need to understand these algorithms to a degree and not apply them blindly in the clinic.

1. Davis KA. Artificial intelligence applications in neurology: Seizing the moment. PL2.001, AAN 2023 Annual Meeting, 22–27 April, Boston, USA.
2. Proix T, et al. Lancet Neurol. 2021;20(2):127–35.
3. Kamousi B, et al. Neurocrit Care. 2021;34(3):908–17.
4. Gill RS, et al. Neurology. 2021;97(16):e1571–e1582.
5. Goldenholz DM, et al. Ann Neurol. 2020;88(3):588–95.
6. Brinkmann B. Seizure forecasting and detection with wearable devices and subcutaneous EEG – outcomes from the My Seizure Gauge trial. PL4.001, AAN 2023 Annual Meeting, 22–27 April, Boston, USA.

Copyright ©2023 Medicom Medical Publishers



Posted on 26 June 202327 March 2024