Abstract first problem-solving program, or expert system, known

Abstract

With changes in medical technologies, healthcare delivery methods, timelines and payment options, adoption of innovative tools to manage patient information and make decisions are becoming important. The need for data mining and decision-making has put artificial intelligence (AI)-enabled solutions at the forefront of the current wave of healthcare revolution. The aim of AI is to enable greater accessibility, understanding correlation and actionability of healthcare information. Nowadays, the capacity to extract information from disparate information sources, analyze large unstructured data sets and ability of natural language processing allow AI systems to tackle challenges in healthcare coordination that previously had no other means of recourse. In this article we intend to trace the journey of application of artificial intelligence in medical science, discuss its current status in clinical diagnostics and throw light on the AI solutions in the anvil with a focus on India.

Keywords: Artificial Intelligence, clinical diagnostics, machine learning, natural language processing, data mining

Background

Artificial intelligence (AI) aims to mimic human cognitive functions. It is bringing a paradigm shift to healthcare, powered by increasing availability of healthcare data and rapid progress of analytics techniques. AI can be applied to various types of healthcare data (structured and unstructured). Popular AI techniques include machine learning methods for structured data, such as the classical support vector machine and neural network, and the modern deep learning, as well as natural language processing for unstructured data1. The rapid commercialization of machine learning and big data has helped bring AI to the forefront of healthcare and life sciences and is set to change how the industry diagnoses and treats disease.

Evolution of AI

The term artificial intelligence was first coined by John McCarthy a young Assistant Professor of Mathematics at Dartmouth College in 1956 when he held the first academic conference on this subject – Dartmouth Summer Research Project on Artificial Intelligence. Research in the 1960s and 1970s produced the first problem-solving program, or expert system, known as Dendral2. While it was designed for applications in organic chemistry, it provided the basis for the subsequent system MYCIN3, considered one of the most significant early uses of artificial intelligence in medicine. Many of the early efforts to apply artificial intelligence methods to real problems, including medical reasoning, had primarily used rule-based systems, reported Duda RO and Shortliffe EH in Science back in 19834. Such programs were typically easy to create, because their knowledge was catalogued in the form of “if/then” rules used in chains of deduction to reach a conclusion. However, MYCIN and other systems such as INTERNIST-1 and CASNET did not achieve routine use by practitioners though4-6. Nevertheless, in many relatively well-constrained domains rule-based programs have begun to show skilled behaviour7. This is true in several narrow domains of medicine as well7,8, but most serious clinical problems are so broad and complex that straightforward attempts to chain together larger sets of rules encounter major difficulties. Essentially rule based models cannot incorporate the myriad biological variations seen both in humans and in diseases. Absence of such logical human reasoning models, which is based on the addition of new rules, leads to unanticipated interactions between rules and thus to serious degradation of program performance9-11.

Given the difficulties encountered with rule-based systems, more recent efforts to use artificial intelligence in medicine have focused on programs organized around models of disease. Efforts to develop such programs have led to substantial progress in our understanding of clinical expertise, in the translation of such expertise into cognitive models, and in the conversion of various models into promising experimental programs. Of equal importance, these programs have been steadily improved through the correction of flaws shown by confronting them with various clinical problems12.

AI can definitely assist physicians to make better clinical decisions or even replace human judgement in certain functional areas of healthcare (e.g., radiology). The increasing availability of healthcare data and rapid development of big data analytic methods have made possible the recent successful applications of AI in healthcare. Guided by relevant clinical questions, powerful AI techniques can unlock clinically relevant information hidden in the massive amount of data, which in turn can assist clinical decision making13-15.

What can AI systems do in Clinical Diagnostics currently?

·         AI can use algorithms to ‘learn’ patterns from a large volume of healthcare data, and then use the obtained insights to assist clinical practice. It can also be equipped with learning and self-correcting abilities to improve its accuracy based on human feedback.

·         An AI system can assist physicians by providing up-to-date medical information from journals, textbooks, and clinical practices to inform proper patient care16.

·         An AI system can help to reduce diagnostic and therapeutic errors that are inevitable in the human clinical practice17,18.

·         An AI system extracts useful information from a large patient population to assist making real-time inferences for health risk alert and health outcome prediction19.

·         An AI system can interpret an individual’s health status based on vital statistics and diagnostic test results.

How is an AI system developed?

Before AI systems can be deployed in healthcare applications, they need to be ‘trained’ through data that are generated from clinical activities, such as screening, diagnosis, treatment assignment and so on, so that they can learn similar groups of subjects, associations between subject features and outcomes of interest. Once ‘trained’ the quality of AI system is often measured by comparing the results to those derived from human experts, a product validation process. However, since the field is ever evolving there are no clear specifications in validation techniques. One way of validating the systems is by using existing data from clinical case reports. These clinical data often exist in but not limited to the form of demographics, medical notes, electronic recordings from medical devices, physical examinations and clinical laboratory and images20. Specifically, in the diagnosis stage, a substantial proportion of the AI literature analyses data from diagnosis imaging, pathology testing, genetic testing and electrodiagnosis.

 

AI devices

The above discussion suggests that AI devices mainly fall into two major categories. The first category includes machine learning (ML) techniques that analyse structured data such as imaging, genetic and EP data. In the medical applications, the ML procedures attempt to cluster patients’ traits or infer the probability of the disease outcomes21. The second category includes natural language processing (NLP) methods that extract information from unstructured data such as clinical notes/ medical journals to supplement and enrich structured medical data. The NLP procedures target at turning texts to machine-readable structured data, which can then be analyzed by ML techniques22.

Few Examples:

Worldwide

·         Tennessee-based molecular diagnostics start-upIQuity will release its IsolateMS blood test, which uses machine learning to recognize differentially expressed protein-coding genes and noncoding-genes, markers of multiple sclerosis.

·         Cognoa, a Palo-Alto-based digital health startup, has developed a machine learning platform capable of diagnosing developmental delays in children utilizing only information and videos provided remotely by parents. It has already evaluated 300,000 children and the company just completed another round of funding in preparation for FDA approvals.

·         Google’s DeepMind, in collaboration with England’s NHS, is in the process of feeding its algorithm over one million digital eye scans. It has already demonstrated the ability to identify sight-threatening conditions with equal accuracy to human ophthalmologists.

·         A research team from Beth Israel Deaconess Medical Centre and Harvard Medical School has trained a neural network to interpret pathology images for tumors. In studies, the network’s diagnostic success rate was 92%, which compared to 96% for the human pathologists that participated. Yet, the result most promising: when combined, the machine and human results reached 99.5% accuracy.

·         Researchers at New York University’s Langone Medical Center have taught a machine learning algorithm to diagnose PTSD only by listening to a person’s speech pattern. Its diagnostic success rate has reached 77%. Similar voice-only approaches have also shown promise detecting Alzheimer’s.

In India

·         Health Vectors based out of Bengaluru has developed an algorithm to conduct disease risk profiling and reduction pathways based AI system that suggests diet, physical activity and lifestyle modifications recommended as per the risk status of an individual.

·         Bengaluru based, SigTuple applies the latest advances in artificial intelligence towards solving the healthcare diagnosis problem. They build algorithms, which learn from medical data, and help physicians by automating disease screening and diagnosis. They are testing these algorithms through low-cost diagnostic devices and a cloud-based intelligent platform. Sigtuple’s product, Shonit automates the procedure of medical diagnosis to reduce the time and effort. Sigtuple’s core product Manthana is an AI driven learning platform that provides solutions for automated analysis of peripheral blood smear, urine and semen sample, retinal scans, and chest x-rays.

·         Based on an individual’s demographics and clinical history another Bengaluru based startup, Healthi, can recommend most relevant diagnostics tests, use AI based algorithms to analyse the test results and recommend if one needs a physician consultation.

·         Bengaluru based AIndra Systems has the vision to build state of the art medical devices for screening Cervical Carcinoma. They have already created CervAstra a combination of their compact walkaway Autostainer ‘Intellistain’, Image Acquisition Device ‘Visionex’, which can digitize slides at 15 mins each and AI Algorithm ‘Astra’, which can analyze wholes slides and segregate cells into High grade squamous intraepithelial lesion (HSIL) , Low grade squamous intraepithelial lesion (LSIL) and Squamous Cell Carcinoma types.

·         Bengaluru based Niramai Health Analytix Pvt Ltd is using a high-resolution thermal sensing device and a cloud-hosted analytics solution for analyzing the thermal images for screening for breast cancer. Their solution uses big data analytics, artificial intelligence and machine learning for reliable, early and accurate breast cancer screening.

·         Gurugram based Chironx AI provides AI-based computer-assisted detection for diagnostic clinical imaging. It develops artificially intelligent machine learning-based computer-assisted detection (CADx) plugins for clinical application in those clinical settings where resources are less and burden is more. They are involved in analyzing images and researching algorithms for diagnosis of respiratory infections and other critical illness and their application for the development of automated CADx tools.

·         Mumbai based Qure.ai is applying tailored deep learning deep learning algorithms to medical images to make the most of the rich 3-dimensional information contained in medical images for producing a diagnosis. One of their key focus areas is algorithm interpretability, ensuring that the reason for a suggested diagnosis is clear to a doctor. Their deep learning algorithms precisely quantify disease and tumor volumes, so that patient response to therapy can be monitored closely.

·         Bengaluru based Touchkin is a predictive healthcare startup. Its machine learning platform identifies the potential health problem through changes in the patterns of communication, activity, and sleep of a person, which are tracked by the smartphone. Touchkin platform. It collects the data from mobile phones and sensors and uses the data to identify changes in behavioural patterns.

 

What lies in near future?

These impressive results are only a precursor to how precise and prolific machine learning’s diagnostic powers will become. As researchers feed the algorithms more data, as the tools begin to be adopted by hospitals and physicians, “teaching sets” will grow and in turn, diagnostic accuracy will improve. The data may come from the Internet of Things (IoT), a network of physical devices and other items, embedded with electronics, software, sensors, and network connectivity, which enables these objects to collect and exchange data23. By 2020, 40% of IoT-related technology is expected to be health-related, more than any other category, making up a $117 billion market24. The convergence of medicine and information technologies, such as medical informatics, will transform healthcare as we know it, curbing costs, reducing inefficiencies, and saving lives25.

As we have discussed in the past, AI has transformative promise across the healthcare landscape, from empowering personalized medicine to improving operational efficiency and predictive cost management. However, diagnosis appears the most-imminent application. The tech giants that dominate the machine learning as a service market—currently IBM, Google, Microsoft, and Amazon as well as start-ups—will be well-positioned as hospitals and diagnostic chains require big data analytics infrastructures.

Lingering questions

However, there remain some substantial questions about the efficacy of integrating machine learning into the diagnostic process. What would be medicolegal ramifications if an algorithm is responsible for a misdiagnosis? Will overdependence on technology blunt intuitive clinical decision making of physicians, compromising the overall outcome of health management and, thus increasing the cost? Should physicians trust an algorithmically-derived diagnosis if they know how a computer system arrived at the conclusion it did? Who will bear the legal responsibility of such report generated by an AI system?

Endnote

Regardless of the still unknown, machine learning as a diagnosis tool has proven its potential to save lives and 2017 appears the tipping point26. Of all the sectors in India, Artificial Intelligence is poised to disrupt Healthcare the most, in the coming years. Through the application of machine learning, data mining, natural language processing (NLP), and advanced analytics, artificial intelligence will assist doctors in diagnosing diseases faster.

Reference

1. Jiang F, Jiang Y, Zhi H, et al. Artificial intelligence in healthcare: past, present and future. Stroke and Vascular Neurology 2017;0:e000101. doi:10.1136/svn-2017-000101

2. Lindsay, R. K., Buchanan, B. G., Feigenbaum, E. A., & Lederberg, J. (1993). DENDRAL: a case study of the first expert system for scientific hypothesis formation. Artificial intelligence, 61(2), 209-261

3. Clancey, W. J., & Shortliffe, E. H. (1984). Readings in medical artificial intelligence: the first decade. Addison-Wesley Longman Publishing Co., Inc.

4. Duda RO, Shortliffe EH. Expert systems research. Science.1983;220:261-8

5. Randolph AM, Melissa AM, Sue MC, Fred EM Jr, and Jack DM. The INTERNIST-1/Quick Medical Reference. Project—Status Report. West J Med. 1986 Dec; 145(6): 816–822.

6. Kulikowski C, Weiss S. Strategies of data base utilization in sequential pattern recognition. Proc IEEE Conf Decision & Control, 11th Symp on Adaptive Processes; 1972:103-5.

7. Buchanan BG. Expert systems: working systems and the research literature. Expert Systems. 1986;3:32-51

8. Buchanan BG, Shortliffe EH, eds. Rule-Based Expert Systems: The MYCIN Experiments of the Stanford Heuristic Programming Project. Reading: Addison-Wesley Publishing Co.; 1984

9. Clancey WJ, Letsinger R. Neomycin: reconfiguring a rule-based expert system. In: Proceedings of the Seventh International Conference on Artificial Intelligence. Los Altos, California: M. Kaufmann Publishers; 1981:829-36

10. Davis R. Expert systems: Where are we? and where do we go from here? AI Magazine. 1982;3:3-22

11. Schwartz WB, Patil RS, Szolovits P. Artificial intelligence in medicine: where do we stand? N Engl J Med. 1987;316:685-8.

12. Peter S, Ramesh SP, and William BS. Artificial Intelligence in Medical Diagnosis. Cambridge and Boston, Massachusetts. Annals of Internal Medicine, January 1988;108(1): 80-87

13. Murdoch TB, Detsky AS. The inevitable application of big data to health care. JAMA 2013;309:1351–2.

14. Kolker E, Özdemir V, Kolker E. How Healthcare can refocus on its Super-Customers (Patients, n=1) and Customers (Doctors and Nurses) by Leveraging Lessons from Amazon, Uber, and Watson. OMICS, 2016;20:329–33

15. Dilsizian SE, Siegel EL. Artificial intelligence in medicine and cardiac imaging: harnessing big data and advanced computing to provide personalized medical diagnosis and treatment. CurrCardiol Rep 2014;16:441

16. Pearson T. How to replicate Watson hardware and systems design for your own use in your basement.  2011 Available from https://www.ibm.com/developerworks/community/blogs/InsideSystemStorage/entry/ibm_watson_how_to_build_your_own_watson_jr_in_your_basement7?lang=en (accessed on Nov 20, 2017)

17. Graber ML, Franklin N, Gordon R. Diagnostic error in internal medicine. Arch Intern Med 2005;165:1493–9

18. Lee CS, Nagy PG, Weaver SJ, et al. Cognitive and system factors contributing to diagnostic errors in radiology. Am J Roentgenol 2013;201:611–7.

19. Neill DB. Using artificial intelligence to improve hospital inpatient care. IEEE IntellSyst 2013;28:92–5

20. Administration UFaD. Guidance for industry: electronic source data in clinical investigations. 2013 Available from https://www.fda. gov/ downloads/drugs/ guidances/ ucm328691.pdf (accessed on 20 Nov 2017)

21. Administration UFaD. Guidance for industry: electronic source data in clinical investigations. 2013 Available from https://www.fda.gov/downloads/drugs/guidances/ucm328691.pdf (accessed on 20 Nov 2017)

22. Murff HJ, FitzHenry F, Matheny ME, et al. Automated identification of postoperative complications within an electronic medical record using natural language processing. JAMA 2011;306:848–55.

23. Zanella A, Bui N, Castellani A, Vangelista L, Zorzi M. Internet of things for smart cities. IEEE Internet Things J. 2014;1(1):22–32.

24. Bauer H, Patel M, Veira J. The Internet of Things: sizing up the opportunity Internet New York (NY): McKinsey & Company; c2016. Available from:http://www.mckinsey.com/industries/high-tech/our-insights/the-internet-of-things-sizing-up-the-opportunity. (accessed on 22 Nov 2017)

25. Dimiter V. Dimitrov.Medical Internet of Things and Big Data in Healthcare.Healthc Inform Res. 2016 Jul; 22(3): 156–163.

26. Artificial Intelligence is on the precipice of revolutionizing medical diagnosis. Available from https://latest.13d.com/artificial-intelligence-is-on-the-precipice-of-revolutionizing-medical-diagnosis-be6427239f58 (accessed on 22 Nov 2017)