Harnessing artificial intelligence to bridge the neurosurgery gap in low-income and middle-income countries

Introduction

Medicine constantly evolves to improve the standard of patient care [1]. Among the many advancements is artificial intelligence (AI), which is the use of technology and computers to simulate intelligent behaviour and critical thinking similar to that of a human [2]. AI incorporates machine learning (ML) algorithms and deep learning networks. It has the ability to collect massive amounts of data and derive complex patterns from it, enhancing patient care with medical management optimisation and error reduction analyses [2]. In general, artificially intelligent systems have played a critical role in medicine by improving diagnostic algorithms (ie, image interpretations), developing efficient and individualised treatment protocols, catalysing drug development, storing electronic medical records and increasing healthcare delivery productivity. As such, AI can reduce personnel demand while cooperatively increasing efficiency and decreasing potential errors [2].

Neurosurgery resides at the nexus between technology and medicine. AI and ML can be applied to neurosurgery, which frequently employs high-tech medical equipment and information systems with complex data. AI can also improve high-resolution radiological imaging to eliminate invasive diagnostic procedures, as well as identify preoperative, perioperative and postoperative complications to better quantify risk factors and improve patient aftercare [2]. AI has the potential to improve diagnostic accuracy and treatment access in low-income and middle-income countries (LMICs). Despite the benefits, several barriers limit AI access in LMICs. A study on neurosurgical access by Punchak et al revealed that out of the approximately 68 LMICs, 11 countries reported no practising neurosurgeons with over 5 million individuals in LMIC not being able to receive essential neurosurgical care annually [1, 2]. These results demonstrate the scarcity of neurosurgical services in LMICs which form a significant portion of the global population.This editorial discusses the state of current neurosurgical practice in LMICs and the role of AI in improving care efficacy, and forecasts how AI integration into neurosurgery can shape the neurosurgical landscape in LMICs.

Current state of neurosurgery practice in LMICs: pitfalls and upgrades

Several barriers that prevent timely neurosurgical access in LMICs are multifactorial and primarily include shortage of physicians, lack of sophisticated equipment required for procedures and relatively high costs [1]. In LMICs, there is a massive shortage of neurosurgeons compared with neurosurgical demand, with an estimated 1 neurosurgeon per 10 million population [3]. As a result, providing essential neurosurgical care is a significant challenge in LMICs. Another contributing factor is inefficiencies in practice caused by gaps in education, lack of instrumentation, and a higher reliance on personal clinical judgement over imaging [3]. These issues will only become exemplified as the demand for neurosurgical care increases, which can contribute to physician burnout in the long term. Studies have identified burnout as a major contributor to medical errors.These could lead to a catastrophic cascade effect of inpatient mortality and postoperative complications leading to long-term disability [3].

The scarcity of data on the current global neurosurgical workforce, availability of equipment and accessibility in LMICs make it difficult to formulate recommendations that would prove beneficial. However, there have been some efforts to improve transparency of neurosurgical care in LMICs and increase access to neurosurgical practices. The World Federation of Neurosurgical Societies (WFNS) training centre was designed to provide partial and full training to more than 58 neurosurgeons [4]. Its success is thought to have caused a ‘ripple effect’, resulting in a fivefold increase in the number of neurosurgeons in sub-Saharan Africa (SSA), from 79 in 1998 to 369 in 2016 [4]. Partnerships between LMICs and high-income countries (HICs) is another strategy to tackle the shortage of neurosurgeons. Prior to 2007, Uganda had no neurosurgery training camps. Duke University Medical Centre in Durham, North Carolina, collaborated with Mulago faculty to create the Uganda East African Training programme, with the goal of training more neurosurgery residents [5].

Current efforts by LMICs to integrate AI in medicine and neurosurgery with the potential to transform patient care

The widespread integration of AI in medicine is due to its potential to improve patient outcomes and complement the skills of doctors. Many different LMICs have embraced the use of AI for its potential benefits. For example, sub-Saharan Africa (SSA) implementing medical decision support system (MDSS), which refers to any mobile electronic device that can provide medical advice for health workers and extend the availability of care into rural areas [6]. Another example is seen in China with the use of a portable all-in-one diagnostic station that automatically upload results and medical records onto a data analysis system, generating diagnoses in a timely fashion.⁶ Other efforts to increase AI accessibility in low-resource settings include the development of widespread mobile phone penetration, advances in cloud computing, significant investments in digitising health data, and the introduction of mobile health applications [7]. Examples include an automated, mobile device-based AI diagnostic system which can analyse Giemsa-stained peripheral blood samples combined with light microscopy images to diagnose malaria with an accuracy of 91% [7]. Developing similar solutions for neurosurgical care can help breakdown accessibility barriers by providing lower-cost diagnostic tools.

AI and ML are both concepts that could benefit neurosurgery, a specialty that is under-represented in LMICs. Many AI health interventions have shown promising preliminary results and will soon be augmented with existing strategies for delivering safe, affordable and timely neurosurgical care in LMICs [2]. Particularly in disease diagnosis, where AI-powered diagnostic techniques can be used in countries with a scarcity of healthcare providers, ML-based tools can be used to supplement clinical knowledge [8]. The rise of AI in big data and ML can assist neurosurgeons in patient care, diagnostic efficiency, treatment algorithms and clinical decision-making, and in handling repetitive work processes without the risk of burnout. AI can work synergistically with neurosurgical practice and improve the efficiency needed to address timely care with clinical precision [6]. A New York based market research firm, Frost and Sullivan estimates that AI has the potential to improve patient outcomes by 30%–40% [8]. Numerous studies link these findings by demonstrating that door-to-needle times for care play a important role in reducing mortality and improving prognoses [7]. AI can be used to decrease the time necessary to receive and interpret brain images that could guide neurotrauma care, as well as other functionalities such as stroke identification and tumour resection [8]. This can improve diagnostic accuracy and provide faster treatment especially in areas such as SSA without an evident physician available to interpret the images [8].

The benefits of AI come from a possible reduction in surgeon workload and bodily stress, reduction in operation time and a reduction in possible surgical errors [7]. This can especially benefit in low-income or rural settings with a smaller healthcare force compared with other higher-income areas. These potential benefits among others warrant further research to allow better AI integration for neurosurgery in LMICs.

Addressing limitations to AI integration in LMICs

Although quite popular, LMICs are not able to fully incorporate AI in neurosurgery. One possible limitation to proper integration of AI is feasibility in terms of infrastructure and affordability. Infrastructure issues such as lack of electricity and/or internet, inadequate transportation and harsh environment limit accessibility to care. Inadequate transportation was cited as another contributing factor towards the lack of timely access to neurosurgical care [9]. Though AI can aid in the proficiency of surgeons and impact patients on a case-by-case basis, political instability and broken healthcare systems make it difficult to implement technological advancements in general and prevent necessary training needed to use AI-based interventions. In addition, costs of AI-based solutions range from as low as$6000 for simple chatbots to as high as millions of dollars [2]. These costs come from data acquisition, hardware and system costs, as well as system maintenance, which requires technical expertise [7]. These costs may be difficult to ascertain in many LMICs especially in neurosurgery, where the cost of treatment can already be high [2].

The aforementioned limitations are centred around the ethical ramifications of AI-based applications; the lack of AI governance, coupled with a fragile infrastructure, can negatively affect data management [10]. One of the main functionalities of AI is storing individual and population-based data, often with sensitive patient information heavily guarded in access and ownership. However, data protection and privacy breaches are at risk without the appropriate governance needed to use AI technology. This raises concerns of community-targeted marginalisation and discrimination based on sociodemographic data [5]. Moreover, the lack of governance can allow industries to commercialise AI-based technology as its popularity in the applications of medicine increases. Such issues are at the forefront of ethics in AI, warranting future research on the cost–benefit of AI implementation for policy makers [5].

Future recommendations for LMICs

While acknowledging the barriers to widespread AI integration in neurosurgery in LMICs, steps could be taken to improve infrastructure and technical training. Training campaigns like the WFNS training centre in Rabat, while excellent, are not widespread enough to address the physician shortage. Also, similar to Duke and Mulago, twinning programmes should be expanded to other LMICs: a facility in an HIC paired with a facility in an LMIC to improve access to specialists, collaborative training and research opportunities has been very effective in increasing access to surgical subspecialties [4].

In efforts to improve patient care in low-resource settings and weak infrastructure countries, the RAD-AID Friendship PACS programme can be developed as an on-site server to facilitate cloud infrastructure for AI applications [11]. The server is configured to be temperature resistant, which is important in certain resource-poor institutions that are not able to create rooms to house normal servers. In 2020, these servers were installed in Nigeria, Loas and two hospitals in Guyana. While these developments are very beneficial, they are not enough.

The lack of equipment can also be resolved with equipment donations from HICS. Most importantly,incorporating AI in education can significantly improve training and access to AI. A noteworthy example is RAD-AID which began integrating AI education into conventional radiology training at LMIC sites. Research focusing on developing cost and user-friendly AI tech can greatly benefit LMICs whose gross national income per capita per year does not exceed$11 905 (upper middle-income economy). Creation of a multilevel medical AI service network can facilitate the data sharing needed for proper functioning of AI at all locations around the world. This can efficiently be done by ensuring that the system takes into account the infrastructural issues in certain areas such as lack of internet or electricity by potentially installing necessary telephone lines and wireless internet when necessary, or using a hand crank generator in locations without electricity.

The development of digital health policies and strategies in LMICs that lack such policies and strategies can allow for more comprehensive implementation of AI and digital health strategies. Humanitarian efforts, as well as political activism and lobbying to generate public and government interest, can significantly drive development of infrastructure, transportation and logistics to allow for better integration of AI in neurosurgery in LMICs, allowing for better medical care and patient outcomes, and thus serving as a cornerstone towards health equity.

Conclusion

Though AI could improve neurosurgical care in LMICs, limitations that hinder these developments include infrastructure, financing and lack of governance. However, efforts such as increasing the number of training programmes for neurosurgeons and provision of infrastructure for increased research in AI are being implemented, which in complement with others can enable increased integration of AI in LMICs and at par healthcare provision.

Twitter

Rohan Yarlagadda @rohanyarla and Precious Peculiar Olatunbosun @Peculiarmed

Contributors

WAA, AM, MK and JK conceptualised the topic and coordinated the reading, writing and editing of the manuscript. JK, AM, RY, MN, MK, A-RT and PPO contributed to various aspects of reading, data collection, writing the original draft and implementing changes for critical revision. AI and VS contributed to reviewing and edits. All authors approved the final draft.

Funding

The authors have not declared a specific grant for this research from any funding agency in the public, commercial or not-for-profit sectors.

Competing interests

None declared.

Provenance and peer review

Not commissioned; externally peer reviewed.

Ethics statements

Patient consent for publication

Not applicable.

Ethics approval

Not applicable.

References