AI is hot

Artificial Intelligence (AI) is a hot topic right now and a rapidly growing area. ChatGPT (a language tool) attracted over one million users in the first five days after it launched late last year. How does AI fit in the context of Traditional East Asian Medicine (TEAM)? How does AI improve the practice of TEAM? How does it change the role of a TEAM practitioner?

What is AI? AI is an overarching name for solving real-world problems using various techniques such as machine learning, deep learning, natural language processing (NLP), robotics, expert systems, and fuzzy logic [1]. Machine Learning is the science of programming machines to process, interpret and analyse data. Deep learning covers advanced machine learning techniques, whereas neural networks help gain insights and solve problems. NLP looks at the human language and aims at gaining insights for better communication with machines. Robotics is the area that creates artificial agents that perform tasks in a particular environment. Expert systems investigate the human decision-making process, and fuzzy logic is the model that looks at degrees of truth rather than linear traditional computer logic.  

So, how do AI branches apply to TEAM? AI relates to a task that a machine could perform independently from a human (a TEAM practitioner), such as determining a TEAM diagnosis based on a patient’s symptoms. But, as many practitioners have their unique style of practice, it cannot be easy to achieve common ground for some of our diagnostic tools, such as pulse or tongue diagnosis.

Research into what treatment components contribute to the treatment outcomes is ongoing. However, acupuncture and Chinese herbal medicine are complex interventions with treatment-specific and non-specific components. Unsurprisingly, it has been found that non-specific treatment components such as the amount of time that a practitioner spends with a patient, attention, education, credibility, expectation, history taking, diagnosis and expectation and palpation of the pulse or abdomen affect the treatment outcome [2]. Palpation is the human touch, and it appears to be an integral part of a patient and a practitioner spending time together to discuss health challenges. It would not be easy to imagine how AI could make up for this invaluable human connection.

TEAM offers popular modalities and is one of the most regularly used complementary therapies in Australia and elsewhere. According to a study, patients seek a practitioner’s assistance, particularly for various chronic conditions, because they desire “supportive, compassionate, safe healthcare” [3]. Unless AI can be programmed to be supportive and compassionate the way a human individual can, the chances for a practitioner to be replaced by AI are unlikely. Regardless of the above, AI is of great assistance to balance some of the shortcomings of TEAM practice today, but I doubt that AI will replace the human touch.

AI to decode TEAM textbooks

One of the challenges that we face as TEAM practitioners is to opt into practice the way our forebearers (before the Chinese revolution) in ancient China did. It appears that to this day, there is a division between the ancient (classical style) and the modernised version of TEAM. Nevertheless, no matter how TEAM practitioners practice in the community, there always seems to be a lingering interest and craving for ancient wisdom. With AI being more commonly introduced into biomedicine [4], there are also efforts to tap into ancient TEAM textbooks, some of which are traditionally carved onto bamboo slips and silk [5].

Managing such vast pre-clinical data poses many challenges, especially with the resources created in the pre-paper eras. As branches of AI make their way into TEAM, methods to study and decode ancient TEAM textbooks have been investigated. Various sequential methods are needed: Firstly, database construction for cognitive and linguistic analysis (semantic analysis). At the same time, deep learning combined with fuzzy logic mathematics to create knowledge graphs, followed by knowledge graph application to decode historically applied diagnosis and treatment principles [6]. Theoretically, the outline of this process is thorough, given that emerging “interpretational” methods and technology can be applied. As I am not an expert in AI or any of the AI branches, it isn’t easy to visualise the outcome of this process. However, my interest is sparked as, like many of my contemporary TEAM colleagues, I am interested in discovering ancient TEAM knowledge to enrich my clinical practice.

The creation of knowledge trees rather than knowledge islands

A challenge is that different schools of thought or models during TEAM’s evolution in empirical China may view diagnosis and treatment approaches diversly. For example, the concept of the six-conformation model or the focus on the spleen and stomach school are, despite being both critical parts of TEAM, two different views or focuses on what may create the imbalance or disease in a patient [7]. The emerging technique of Knowledge Graph aims to connect “fragmented pieces of knowledge” [8] to overcome knowledge islands and place them on a knowledge tree. In this way, connections and relationships for individual parts can be established and suitably develop clinical practice. Knowledge graphs are built based on databases that already exist. Technology and scripts are then set to produce the visualisation of the data, enable data retrieval, and make tailored recommendations for the patient in care. Irreversibly, it still requires the TEAM practitioner to ask the right questions, note details on the case report, take the pulse, and examine or listen to the type of cough the person presents with. Astonishingly, though, the TEAM practitioner could access a solid knowledge base that AI produced to understand graph mechanisms.  

Latent tree models

Compared to the Western medical system, TEAM has frameworks that emerged over many years of evolution based on many other influences such as cultural, societal, religious, and environmental. Examples are the Yin-Yang principles, the five phases model, the organ systems, the mind-body relationship model, channels and collaterals, qi, blood and fluids and diagnosis and treatment of syndromes. The syndrome differentiation is a crucial aspect of TEAM and links the previous models, so syndrome differentiation becomes the overarching mechanism of TEAM [9]. But this vital area of TEAM causes the most scrutiny because of the diversity of views of a practitioner based on their education and focus.

Syndrome differentiation is gathering patient data by inspection, interrogation, palpation, auscultation, and olfaction and then analysing and categorising the different components into a syndrome or several syndromes. Still, this data might be prone to subjectivity. The latent tree model technology aims to use machine learning to investigate and systematically collect patient data and perform a cluster analysis. This method seems to be based on mathematical calculations [10]. If this method proves feasible, it will significantly assist practitioners in achieving a more congruent approach to agree on pattern differentiation, improving patient-centred care. However, TEAM integration into biomedicine should be considered carefully as there are significant, incompatible differences [9].

AI to overcome subjectivity in TEAM pulse and diagnosis

From my understanding, reliable pulse or tongue diagnosis still requires the assessment of a TEAM practitioner with experience in this method. Only some components of pulse diagnosis can be substituted by AI and rely on classic machine-learning algorithms. I am unfamiliar with the individual techniques, but those methods don’t seem comprehensive, hence the requirement of data needing to be gauged by humans.

It has also been suggested that deep learning is the next step in achieving AI-driven pulse and tongue diagnosis with a convolutional neural network (CNN), which has shown higher accuracy rates in pulse diagnosis [11]. CNN has matured since the 1960s and is recognised as excellent at pattern recognition. CNN’s adaptability and use in different disciplines could investigate human pulse signals [12]. In the evolution of pulse diagnosis in TEAM, AI’s pulse reading would improve the diagnosis because the machine reading reduces subjectivity. Subjectivity is vastly present when a human reads the pulse because many factors (temperature, noise, touch, mental state) could influence the pulse reading result.

Similarly, to pulse diagnosis, AI can assist in tongue diagnosis, a key feature in TEAM. It has been found that the best method to standardise tongue diagnosis is using CNN. In contrast, several steps are involved in categorising the image of a tongue with over 90% accuracy of categorisation. The use of CNN for tongue diagnosis is claimed to predict the early detection of certain conditions [11]. If a practitioner opts for tongue diagnosis, environmental conditions (such as light) are essential to determine the colour of the tongue body and the coating. These features are the most important for pattern differentiation and successful treatment plans.

AI to strengthen diagnosis in TEAM

The basis of data mining is that data exists in the first place. Data also needs to be structured so it can be mined. As previously mentioned, TEAM data hardly exists in countries where there is a separation between the primary health care system and TEAM, unlike in China. Indeed, none of the data is stored in a repository. Zhang et al. looked at a diagnostic system to enhance the diagnosis and prediction of common TEAM illnesses by assessing several thousand patient records from a hospital in China. Their approach involved four different subsystems based on various algorithms to complete the task [13]. My limited understanding of computation, mathematics, statistical models, and contemporary methods of such investigations only allows me to comprehend the mammoth task and resources behind such a project. But, once again, my interest is sparked by what type of details this amount of data could provide and how it could further develop my clinical skills.

Conclusion:

Is the most considerable apprehension towards AI perhaps that we fear that AI has the potential to take over and control and rule our lives? I certainly have those hesitations towards technology. The power and velocity of data processing have long outperformed the human brain, and I doubt that the advancement of AI can be stopped now. As humans, it is our responsibility to create ethical standards [14] incorporating the use and development of AI that will not disadvantage any individual.

Is machine learning and AI beneficial to the improved practice of TEAM? Based on my assessment of how areas of TEAM can be enhanced with the assistance of AI and machine learning, I am convinced that the support of AI for TEAM is precious. Developing critical components such as tongue and pulse diagnosis is an exciting venture. Can AI replace the presence of a TEAM practitioner in the consulting room? A human being who will answer questions relevantly express empathy, hope, and urge to help the best they can sometimes is all our patients need to start the journey of recovery, so I would have to say, no way!

Lastly, the question about how AI will impact the role of a TEAM practitioner. Given the need for AI to access computer data and algorithms, I imagine digitisation would be apprehensive by some TEAM colleagues, as my findings in semi-structured interviews revealed that 25% of practitioners prefer to handwrite their case notes. However, as I now understand that handwritten free text notes can be included in deep learning or expert systems technology, I can imagine that our colleagues would be more open towards creating the required database for AI to become feasible.

TEAM has always been a medical system based on naturalistic principles, methods and paradigms that reflect the dynamic cycles of life. Thus, I wonder if AI can be adapted to reflect the profoundly natural laws of TEAM. However, I can imagine that some AI could make our medicine potentially more accurate and less subjective.

References:

1.         Jackson, P.C., Introduction to artificial intelligence. 2019: Courier Dover Publications.

2.         Paterson, C. and N. Britten, The patient’s experience of holistic care: insights from acupuncture research. Chronic illness, 2008. 4(4): p. 264-277.

3.         Foley, H., A. Steel, and J. Adams, Consultation with complementary medicine practitioners by individuals with chronic conditions: Characteristics and reasons for consultation in Australian clinical settings. Health & social care in the community, 2021. 29(1): p. 91-103.

4.         Maojo, V. and J. Crespo, Challenges for future intelligent systems in biomedicine. Journal of Intelligent & Fuzzy Systems, 2002. 12(1): p. 1-3.

5.         Zhang, B.-R., et al., Research on Unearthed Traditional Chinese Medicine Documents. Chinese Medicine and Culture, 2021. 4(2): p. 114.

6.         Gao, L., C.-H. Jia, and W. Wang, Recent advances in the study of ancient books on traditional Chinese medicine. World Journal of Traditional Chinese Medicine, 2020. 6(1): p. 61-66.

7.         Wiseman, N., Traditional Chinese medicine: a brief outline. Journal of chemical information and computer sciences, 2002. 42(3): p. 445-455.

8.         Yu, T., et al., Knowledge graph for TCM health preservation: Design, construction, and applications. Artificial intelligence in medicine, 2017. 77: p. 48-52.

9.         Hai, H., Kuhn and the two cultures of Western and Chinese medicine. 2009.

10.       Zhang, N.L., et al., Latent tree models and diagnosis in traditional Chinese medicine. Artificial Intelligence in Medicine, 2008. 42(3): p. 229-245.

11.       Wang, Y., et al., The Impact of Artificial Intelligence on Traditional Chinese Medicine. Am J Chin Med, 2021. 49(6): p. 1297-1314.

12.       Zhang, S. and Q. Sun. Human Pulse Recognition Based on Convolutional Neural Networks. in 2016 International Symposium on Computer, Consumer and Control (IS3C). 2016.

13.       Zhang, H., et al., Artificial Intelligence–Based Traditional Chinese Medicine Assistive Diagnostic System: Validation Study. JMIR Med Inform, 2020. 8(6): p. e17608.

14.       Hamet, P. and J. Tremblay, Artificial intelligence in medicine. Metabolism, 2017. 69: p. S36-S40.


Leave a Reply

Your email address will not be published.