The field of artificial intelligence (AI) and machine learning (ML) has rapidly emerged as a force in clinical medicine and is expected to continue evolving in the future. AI and ML have the potential to revolutionize the delivery of healthcare, facilitate the design and conduct of clinical trials, and enhance patient outcomes. However, adoption of AI and ML applications has varied across different fields of medical science. Significant challenges related to the need for real-time diagnosis and management of patients are especially relevant issues in critical care and neurocritical care, with other limiting factors including pre-existing biases in original data sources, implementation bias while applying AI recommendations, and the “black-box” nature of many AI and ML algorithms. In low- and middle-income countries (LMICs) whose resources are limited, the application of AI and ML is further affected by factors like limited funding, lack of proper infrastructure, lack of technical expertise, and limited access to data. As a result of these unique local factors, algorithms initially trained in higher-income countries may not be generalizable to LMICs.

Here we gather the perspectives of national representatives from the Neurocritical Care Society’s (NCS) Asia Oceania chapter, each of which consider the current understanding, applications, and perceived barriers for implementation of AI and ML in neurocritical care in this region, as well as potential ways forward. This overview can help form the basis for future collaborations and larger-scale research to enhance effective utilization of AI and ML in this region.

Australia

As a resource rich country, Australia has access to the required technology and expertise to implement AI in healthcare. Indeed, the use of AI and machine-learning to identify signals not immediately discernible using standard analytics is becoming increasingly common. This can range from real-time AI-assisted decision-support to ML-based prognostic models that can provide accurate estimates of the likely clinical trajectory or outcome for a given patient. However, to date, these approaches have been limited to research settings due to a number of concerns. The main factors limiting the more widespread use of AI/ML relate to the following:

• Regulatory and governance requirements that limit the sharing of data between regions and healthcare providers, which in turn limits the development and validation of AI/ML analytics and outputs
• The absence of datasets of suitable scale and granularity, which leads to the use of data that are often not representative of the Australian population
• Clinician anxiety about the reliability, accuracy, and validity of AI/ML derived outputs
• Medico-legal concerns and the dehumanization of healthcare

Keys to successfully implementing AI/ML analytics in neurocritical care in Australia include greater centralized collection of highly granular physiological, treatment, and outcome data; improved access to data repositories and more efficient data linkage, automated data pipelines, cleaning, and curation; infrastructure modernization; increased collaboration between clinicians and AI/ML experts; and support from regional and federal government agencies and funders. In addition, and perhaps most importantly, the implementation of AI in neurocritical care requires ongoing engagement with consumers and clinicians to ensure the outputs generated are patient-focused and meaningful to those with a lived experience in neurocritical care.  Moreover, any AI/ML derived interventions will require robust assessment in randomized clinical trials.

India

In recent years, we have observed remarkable expansion and advancement in the deployment of AI in healthcare, and its application in neurocritical care has also escalated. Neurocritical care is a demanding field, requiring constant vigilance and rapid decision-making to manage patients with acute neurological conditions. AI has emerged as a promising tool to assist neurocritical care clinicians by facilitating early prediction of neurological deterioration and enhancing management and ultimate outcomes, and we have already begun implementing AI here for these reasons.

Intracranial pressure (ICP) is among the most widely monitored parameters in neurocritical care given its correlation with mortality and other outcomes. Utilizing AI has enabled continuous ICP monitoring, which can enhance a clinician’s ability to promptly address undesirable changes and refine treatment protocols. AI can also aid in the early detection of delirium, the prediction of acute kidney injury, the identification of early seizures after intracerebral hemorrhage, and the overall assessment of seizure risk. From a stroke management perspective, AI has the potential to improve early detection and prognostication, identify high-risk patients, and optimize treatment to reduce morbidity and mortality. Across ICU settings, AI-driven smart pumps have helped optimize medication titration, outperformed conventional scoring systems like APACHE and SOFA in predicting early mortality, aided in brain death confirmation, and predicted sepsis and infections (e.g., CLABSI and C. difficile infections).

Despite its promise, AI poses challenges such as the risk of misdiagnosis, lack of regulatory frameworks, and medico-legal concerns, while questions of accountability, patient autonomy, and the doctor-patient relationship remain unresolved. Addressing these issues is essential for AI’s safe and ethical integration into neurocritical care.

Japan

AI is becoming a welcome tool for the standardization and optimization of clinical practice, especially in rural areas. However, current limitations include the following aspects:

• Patients with communication barriers (e.g., those in a comatose state)
• New patients in a first encounter before an established diagnosis
• Promoting and motivating patient education and self-care
• Medico-legal issues
• Building rapport with patients
• Mental health care

Although Japan is one of the leading countries in the development of AI and ML​​, AI use in the clinical setting remains insufficient. Japan’s utilization of AI is high in image analysis (about 20% in all facilities and about 30% in university hospitals), followed by genome medicine, diagnostic and treatment support, surgical  procedures, nursing care for patients with dementia, and drug discovery. However, Japan has disparities in the prevalence​​ of electronic health records​​ and ordering systems—the basis for AI in healthcare—which are present in 95% of hospitals with more than 400 beds but only about 60% in small hospitals and clinics

Multilingual AI models are essential for healthcare, and such models have already been produced in Japan. However, in addition to differences in language, AI models should ideally consider other differences among individuals and populations including personality, religion, economic status, and so on. NCS and its regional chapters, with their wealth of global and multi-disciplinary specialists, are ideal platforms to discuss the incorporation of AI into neurocritical care. With its robust background and potential to promote AI technology, Japan is keen to contribute and collaborate in the discussion.  

Nepal

Although our use of AI and ML in critical care is expanding, critical care clinicians in Nepal have a variable level of knowledge about these technologies. Application of AI and ML has largely been limited to small scale research projects predominantly within the field of radiology, along with the use of some built-in AI features in imaging modalities like ultrasound. Within neurocritical care specifically, clinicians’ awareness and perspectives about AI and ML remain unclear.

The major barriers we face are a lack of knowledge among health care workers, a lack of relevant expertise, a lack of capacity to implement AI and ML (including primitive electronic medical recording systems and electronic databases at most centers), a lack of legislation and policy for adapting and implementing AI, and a lack of involvement from policy makers and funding bodies. We need to work to increase awareness among our health care workers and to identify gaps, so that future plans can build towards meeting the needs of our healthcare landscape. Further implementation will need education of health care workers, building infrastructure with increased capacity, more investment from government and policy makers, AI models that are tailored to local needs, and collaborative efforts both locally and at the international level.

Singapore

The potential for AI in revolutionizing healthcare is untapped. While Singapore has adopted technology in some aspects of health care (e.g., electronic medical patient records), we have an unmet need for amalgamating big data into AI algorithms that allow for the alignment of best care practices across the nation.

Some of the challenges we face include access to a national database health record and the spread of Singapore’s population across different regions with different care needs (e.g., central Singapore has an older population, while the new peripheral regions have a younger population). On an individual level, how do we manage scenarios when clinical management differs from an AI-driven pathway? Would there be medico-legal consequences even if a patient would not do well in either situation?

We would need a consolidated approach to get an AI model that works for Singapore’s unique and rapidly aging population, then validating this model across the country’s populations, regions, and hospitals—an approach that requires collaborations between local experts, hospitals, and agencies to navigate a complex and rapidly changing AI landscape. If these challenges can be overcome, healthcare could indeed be transformed by the adoption of AI in the very near future.

Conclusion

The NCS Asia Oceania chapter and its individual member nations have varying perspectives related to AI and ML. While there is consensus on the need for adapting AI and ML for neurocritical care applications and their potential to improve outcomes, individual nations seem to be at variable stages of adapting and implementing these technologies. Those that have more limited resources appear to be at an early phase, focusing on improving awareness and building capacity, while those with more resources have advanced to work toward overcoming challenges like medico-legal issues, data sharing, and AI tailored to local settings. Well-designed surveys are needed in order to elucidate real-world issues and explore current gaps and barriers related to using AI and ML, while collaborative efforts between member nations and NCS as a whole can help promote the implementation of these technologies, especially in nations with limited resources.

Note: The full list of authors includes Gentle S Shrestha, MD, FNCS; Andrew Udy, PhD, FCICM; Girija P Rath, MD, DM; Masao Nagayama, MD, PhD; Prashant Kumar, MD, MBA, FICCM; Saurabh Anand, MBBS, MD; Yu-Lin Wong, MMed Anaes, ICM, ANZCA; Gene Sung, MD, MPH; and Kapil Zirpe, MD, CHEST, FICCM, FSNCC