Role of Artificial Intelligence (AI) in Neuromuscular & Electrodiagnostic Medicine

AI’s foray into medicine can be traced back to a patient interactive system, ELIZA, which was an early natural language processing computer program developed at MIT in the mid-1960s, but further development was limited by available computing power​ ³​.  There have been significant strides over the past decade, and much more advanced systems have emerged, including GPT-4 which was introduced in 2023. That system can accurately answer most questions on the medical licensing exam, correctly answer most questions from patients or physicians, and could have done the initial draft of this paper. The system “self-trains” by accessing data available on the internet, online textbooks, websites, podcasts etc. Other systems being developed and becoming available have the potential to “listen” to the conversation between a patient and provider and produce a draft of the clinical note​⁴​, and even possibly cue up orders for providers to review before signing. 

Physician/Provider Aspects

As much as the application of AI to healthcare is exciting and promising, providers must exercise caution with full awareness of the limitations and risks involved. Among such considerations is the responsible development of rigorously validated AI technologies which requires that models are trained adequately, transparently and with sufficiently diverse populations to reflect those to whom the technologies will be applied. This is necessary to mitigate the potential disadvantages and harms to underrepresented groups, including minorities. With the rapid pace of development in the field of AI in general, and in healthcare in particular, regulatory measures are significantly lagging, and it is imperative that this limitation is curtailed in the interest of patient safety. Related are novel challenges with ethical and legal issues pertinent to patient privacy (including cybersecurity), and liability for medical errors in which there may have been AI over-reliance. Accordingly, it is important that AI should serve as a clinical decision support tool rather than a replacement for human expert judgement. Providers should be adequately trained in the clinical application of AI technologies and continuously monitor the performance of these to ensure optimal efficacy and risk minimization.​⁵​  

Patient/Public Aspects

AI and patient care: AI can assist in gathering medical data to assist in accurate and timely diagnoses. AI can recognize patterns in NM medicine and identify early signs of diseases that might be overlooked by human physicians It can also help physicians implement early preventative measures.to avoid complications, thereby leading to better health outcomes. ^1,2 

AI and patient management using evidence-based medicine: AI can assist in identifying and providing evidence-based treatment recommendations considering unique patient related factors. It can support clinical decision-making in conjunction with human judgment and thus help optimize efficacy and patient safety. ¹ 

AI role in administrative processes and patient experience: AI can facilitate patient scheduling, as well as assist with coding, billing, prior authorizations and appeal letters for insurance coverage denials. This can lead to improved patient-provider interactions including “liberated” time that could be used for more face-to-face time during encounters, thereby promoting overall better patient experiences. 

AI and ethical considerations in patients’ privacy and safety: Prior to implementing AI in patient care, careful consideration must be given and strict regulations enacted to protect patients’ privacy and ensure data security. Furthermore, the same caution which would be prudent for any new medical technology must be exercised and rigorous oversight will be needed prior to and throughout implementation of any AI programs in healthcare. Some major areas of concern specific to AI in healthcare include: 

• Informed Consent: AI programs may have access to or utilize patient information that the patient has not actually consented to releasing. Care must be taken to understand what information is being used and how it is being used so that patients can be properly informed in an easy-to-understand format and, thus, be able to provide appropriate consent. ³ 

• Bias and discrimination: AI may base consideration of information and/or recommendations on well-known (or yet-to-be discovered) biases that currently exist in available data sets used for the AI model training. ³ 

• Human autonomy: There must be human oversight of any AI programs used in patient care to ensure that human values and ethical considerations are protected and included in any AI-assisted medical decision-making. ² 

The World Health Organization (WHO) has identified 6 core principles to consider when utilizing AI in healthcare¹⁰: 

1. Protect autonomy; 

2. Promote human well-being, human safety, and the public interest; 

3. Ensure transparency, explainability, and intelligibility; 

4. Foster responsibility and accountability; 

5. Ensure inclusiveness and equity; and 

6. Promote AI that is responsive and sustainable. 

Working with Industry, and Regulatory Aspects of AI in EDX and NM Medicine 

As previously alluded to, with the very rapid development in this complex technical area, regulatory bodies have struggled to keep up and provide the necessary guardrails for AI applications in medicine.  

The US Food and Drug Administration (FDA), for instance, has imposed rigorous requirements for medical devices and algorithm licensing. ^11,12 In response to the evolving nature of technologies, the FDA has introduced a new regulatory strategy called the Total Product Lifecycle (TPLC), which allows for continuous modifications of algorithms as they develop and learning occurs over time, enabling effective regulation of Software as Medical Devices (SaMD) throughout their lifecycle. 

The FDA's regulatory process for algorithms comprises three licensing pathways: the Pre-Market Approval pathway, involving a comprehensive review of the algorithm's safety and efficacy before clinical use; the 510(k) pathway, which involves a comparison study with other similar devices/algorithms; and finally, the De Novo pathway, where safety and efficacy are studied for algorithms without comparable devices. ¹¹ To date, the FDA approved over 500 AI/ML-enabled medical devices, with some of these devices utilizing electromyography signals to aid prosthetic movements. However, the focus of most approved algorithms are the fields of radiology and cardiology applications, with only few examples notable for neurology and NM medicine, in particular. 

Despite the relatively slow FDA approval process, there remains a significant lack of a robust framework for assessing and regulating algorithms applied in the medical field, particularly in neurology. A recent systematic review of 53 algorithms with applications in the central nervous system revealed a concerning lack of substantial peer review, with only 10 algorithms being published, and a scarcity of rigorous clinical efficacy studies, often involving biased study populations. ¹³ These findings underscore the importance of advocating for and actively participating in the creation of regulatory frameworks for the use of AI/ML in neurology and NM medicine. This is particularly important in preventing unforeseen and unintentional harms which may arise from the premature deployment of healthcare AI applications which may have had suboptimal development (including inadequate AI model training) and pre-market testing.