ARTIFICIAL INTELLIGENCE & OPTHAMOLOGY ABSTRACT

Transcription

1 ABSTRACT ARTIFICIAL INTELLIGENCE & Artificial Intelligence is an upcoming tool with lofty potentials in all branches of science and technology. The utility of the system has disseminated into the field of medicine. AI tools in ophthalmology have created gadgets of automated diagnosis that identify disease and therapeutic markers to diagnose and treat diverse ocular diseases. The boon and bane of the system has yet to be unravelled. OPTHAMOLOGY Disruptive Diagnostic Techniques in the New Millennium.

2 What is Artificial Intelligence? Artificial Intelligence (AI) is a burgeoning area of computer technology which structures the creation of intelligent machines that mimic human acumen. It is essentially the ability of a mechanized system to exhibit cognitive abilities and thinking generally pertaining to human beings only. Similar to human brain development, these AI systems need to be fed with a set of functional programs which allow them to first learn simple targets like learning and performing a specific task. Following the learning session the systems are allowed to evolve to attain proficiency and competence. This learning is further compounded by use of complex mathematical equations for the system to understand nonlinear relationships between different variables through an information flow referred to as neural networks. This form of higher training allows AI to judge and weigh possibilities of different outcomes, much like an ideal physician or human being. The ability of complex computing to be able to perform pattern recognition by creating complex relationships based on input data and then comparing it with performance standards is a big step forward from just a complex database analysis. History of Artificial Intelligence Artificial Intelligence made its way into technology around the 1950s. John McCarthy, a pioneer technological scientist is known widely as the Father of Artificial Intelligence. He named the technology as Artificial Intelligence at the Conference of Dartmouth. Soon the utility of the tool was sensed everywhere. McCarty developed the LISP programming language, which was regarded as the first program in Machine learning. The AI performance The performance of AI systems has soared to whopping heights in almost all domains including sports, medicine, finance and education. In the year 1996, the then software giant IBM s AI chess-playing computer, Deep Blue defeated the 1

3 chess world champion Kasparov by a score of 4-2. Similarly, different versions of Alpha Go AI computer programs developed by Google DeepMind officially won the board game Go. This achievement of AI technology was acknowledged as one of the Breakthrough of the Year runner-up by the popular research journal Science in But the potentiality and competence of the technology was fully realised after the IBM s AI computer Watson played a face to face quiz show Jeopardy with human opponents. Watson stole the show by defeating Jeopardy champions Brad Rutter and Ken Jennings. These accomplishments help AI to learn and master human complexities like strategic planning, decision making, prediction and negotiation. This strategy can be applied to solve a myriad of complex problems in various fields including medicine and healthcare. The AI systems may uncover patterns that are humanly impossible to uncover due to their rapidity to find associations between variables that are beyond the human brain s capability today. Many of these technological advances are in part due to software that have now been made available by resources related to information and technology. A few examples of commercially available machine learning software include Scikit-learn, Microsoft search systems, IBM Watson, Big ML etc which require computer languages like Python as a platform for working. The recent modalities of AI in other expanses of technology like unmanned copters, automated vehicles, face recognizing tools have revolutionized the society s lifestyle. Some of the activities where AI tools are helpful include: Recognition of speech Learning Planning Problem solving Deep learning and Neural network The efficiency and sensitivity of the human brain is a wonderment in itself. Therefore, mimicking the function of the human brain has been one of the 2

4 fascinating dreams of technologists for decades. The human brain is exposed to millions of sensory data every minute and captures these data in the human memory in a succinct order. Deep learning or hierachial learning is a subset of machine learning system where the programs and algorithms have a structural and functional correlation to mimic the neurons of the brain. The algorithms that are used to train the systems are built on a mathematical model of sample data and gradually train them to solve big data. Since the network of data are composed of a mesh of circuits that closely resemble the neurons of the brain, the system is named as such. The applications of neural network have extended to varied domains including speech recognition, image analysis, adaptive control, computer games, robotic service and the like. Industries which handle large amount of data have adopted AI s technologies like Machine learning, Deep learning and neural networks. AI in healthcare AI has revolutionized the domain of traditional healthcare. The roots of AI have already gone deep into different medical specialities like radiology, dermatology, pathology, drug discovery and ophthalmology in the recent past. The CB Insights 2016 report revealed that about 86% of companies belonging to healthcare, life science and medical diagnostic domains utilize the AI technology. Some of the areas which need AI technology s assistance in todays healthcare world include: Health record and data management - Employ artificial intelligence and digital automation to collect, store, format and trace bulk patient data. Robotic assistance in doing routine tasks - Critical domains like radiology and cardiology require repetitive tasks like test analysis, scans and X-rays. Assistance of AI technology in this area has increased in the recent years Treatment design - AI systems assist in designing the treatment path by studying the patient s health history and disease prognosis. With more than a million new cases of cancer every year, it has become extremely difficult for oncologists to study patients omics data and take critical medical decisions. Many AI projects have come handy for oncologists like the project Hanover 3

5 which assists oncologists in giving precise medical advice supported by cloudbased health analytics. Similiarly personalised drug combinations have been structured for treating different blood cancers based on the genomic status of the patient using deep machine learning techniques. Virtual Consultation and Virtual nurses - Digital applications like Babylon have launched their AI doctor to consult patients through natural language and offer medical advice. Similiarly AI startup Sensely has developed a virtual nurse called Molly to monitor patient s health condition between doctor visits. The Jeopardy famous Watson AI computer has been reframed by IBM to help oncologists offer individualised and evidence based treatment options. Drug creation - AI has offered high throughput screening in designing and remodelling novel drugs for varied diseases. This could save pharma companies and scientists many years of wet lab research on drug discovery. In addition, AI scientists have converted text into organized databases and have built genome-scale knowledge bases to read a myriad of medical articles. Early diagnosis - AI systems have been employed to discover mutations through body scans aiding in early diagnosis of different types of cancers and vascular diseases. Google s Deep Mind technology has been applying cutting edge technologies in AI to study the 3-D model of protein structure which can solve tons of challenges in biology. Similarly dermatologists have used AI technology in tracing melanoma cells to diagnose skin cancer. An AI team of scientists from DeepMind (Google LLC) in association with researchers at Moorfields Eye Hospital, London have invented smart machines which identify even the slightest symptom of macular degeneration and other eye diseases using optical coherence tomography scans. Health monitoring devices - AI s healthcare commodities like wearble sensors and diagnostic devices use data to gauge a patient s health condition in real time. Wearable heath trackers like FitBit, Apple, Garmin etc monitor the vital signs and send alert signals in case of an emergency. 4

6 These innovations are just the tip of the iceberg. Many more are to come as innovation nudges the potentialities of AI and automation. AI in ophthalmology Ophthalmologists face practical impediments with manual examinations which are imprecise and subjective. Like any other healthcare sector, AI has a rousing welcome in Opthalmology. The use of AI systems in ophthalmology helps professionals in diagnosing ocular diseases by gauging the ocular parameters with automated image analysis. So far about two hundred and forty three research articles pertaining to AI application in ophthalmology have been published. Glaucoma, Diabetic retinopathy, Age related macular degeneration and cataract are studied intensely by AI scientists. Some of the imaging modalities like fundus image, optical coherence tomography, ocular ultrasound B-scan and slit-lamp image aid in diagnosing retinal and anterior segment diseases and also evaluate the condition of lens, vitreous, retina, optic nerve and tumor. Some of the long-time goals of AI system in ophthalmology are as follows: To ease the physician s exertion during diagnosis or screening and provide ophthalmological guidance through AI s deep learning and machine learning modalities. To technologically update ophthalmological tools and provide better patient care. To empower hospitals and ophthalmologists to integrate AI tools in their routine clinical practice. Ophthalmologists have been foreseeing myriad of novel automated tools to treat different ocular diseases. Technological giants like Google and IBM have gradually shifted their attention towards the functionality of these AI automated tools in ophthalmology and healthcare. The logical reason behind this endeavour is because the retina of the eye reflects many systemic diseases. So, analysing the retinal structure can predict the patient s overall health condition. Google in association 5

7 with a start up called Verily has launched this initiative in India to detect diabetic retinopathy with excellent diagnostic capability in par with the top opthalmologists and retinal specialists. With AI s assistance, treatment and screening time is greatly reduced so that opthalmologists can consult more patients in less time. The Future of AI in Ophthalmology The advent of Machine learning and AI in the healthcare and medical industry signifies fast and precise diagnosis like never before. The services of AI can be employed in areas where specialist doctors may not be available. The need of automated detection systems is felt more in developing countries where human skills are still exiguous. Diagnosis for ocular ailments like diabetic retinopathy and even glaucoma is highly reliable and accurate in comparison to other conventional ophthalmologic techniques. An integration of the AI based Ophthalmologic Computer Application with the Patient Record Management system will yield deeper insights about the patient. This will help the Ophthalmologist arrive at a holistic assessment of the patient s other disorders. These breakthroughs of AI and machine learning in diagnosis and therapy have warranted potential transformation of therapeutic techniques and drug discovery procedures in the near future. AI scientists and medical professionals claim that deep machine learning can soon be applied into other cardinal areas like cardiology and oncology. However, many experts raise their concern over the reliability of these systems especially in the healthcare domain where accurate diagnosis is inevitable. The medicos would have to learn the technology of the system in order to correlate with their medical practice. Challenges and pitfalls of AI Although AI s advent into medical research is regarded as a meritorious favour of the tech world, the efficacy of AI treatment modalities is still dubious. When tech 6

8 scientists launch their highly commendable AI software to track neoplastic lesions, the new laws and policies give way to ethical commotion. The safety of these so called smart health tech systems is still dodgy. The technology has grown at a cracking pace that it can replace human judgement and take medical decisions. This is where the trouble can start. real harm to patients through unsafe and unregulated systems. AI systems in software or other engineering niches perform under the unwritten dogma Efficiency first; safety next which is completely contrary to medical or healthcare dogmas which emphasis on safety first and performance secondary. Some flaws of AI systems in tracking human or wildlife have already been observed and critically analysed. For instance, the Google photo tracking system identifying people with dark complexion as gorillas has been accused of causing racial commotion. Experts are convinced that fully automated AI systems capable of taking medical decisions can cause more harm than benefits. Challenges in Adopting AI in Ophthalmology The soaring interest in adopting AI as a diagnostic and therapeutic tool in ophthalmology is obvious in today s sitch. Nevertheless, some of the grave challenges of AI systems need to be addressed initially before adopting the system. The credibility and reliability of standard datasets The standard set of images employed as AI s tools has to be comprehensive enough to produce accurate outcomes. Dr Chiang, an ophthalmologist and AI scientist at Oregon Science University, Portland raised suspicions about the reliability of the reference standard image sets used in the software. It is a priority to set a secure reference standard that are common across various people groups in different countries. It is also important to use datasets that everyone agrees and validates. Regulatory crisis Implementation of AI s technology in health care is still a challenge in real-life. Regulations to standardize the safety and efficacy of these health-related AI systems are not well structured. In 2017, FDA issued its first regulatory guideline to assess 7

9 the safety of the system. These modalities are recognized as general wellness products and the regulations for wellness products are not as stringent as other therapeutic or diagnostic tools. Another guideline allows real-world evidence to generalize the efficacy and credibility of these tools, which calls for more clinical trials. This can pose more serious problems without validating the safety of the system. In addition, many laws to regulate Medical Record Management Systems, Data Privacy and Protection have to be amended or modified to protect basic patient and user rights. There are strict regulations in place like the HIPAA and AI based Solutions are expected to conform to HIPAA at a modular level and as a part of an integrated system. Data exchange difficulties Data exchange is another practical impediment in AI healthcare systems. AI systems need to be fed continuously with data from clinical trials. Massive data sharing systems are needed to exchange data on a large scale. This type of data sharing is underway in the US. Bias and false negativity The main advantage of AI over manual labour is minimal bias and manual error. But the system also is prone to commit some grave errors when they adopt and reinforce the biased data fed into the system. The House of Lords Committee on AI reflected that the data sets used to train the system are not representing a wider population which will lead to false analysis and pose potential diagnostic and therapeutic complications. Also, the AI systems are prone to fail in regions were quintessential data is unavailable or scarce. Data safety Several criticisms on data safety have been raised since the conception of the AI ideology. Many professionals have warned potential public backlash at many of AI s healthcare tools. Since all of the personal data in the AI tools is owned by the commercial companies, possibility of data theft and commercialization of patient s personal data can also be suspected. 8

10 Some of the inevitable questions below have to be addressed in the first place before accepting the technological indulgences of the system. Does the system serve only the elites - will it lead to democratic inequality or promote biological superiority? Will the system meddle with patient s privacy or make room for data theft? Can the system be accountable and credible? - When lives are at risk? Can the black box algorithms of the system be credible and explainable? Will the system replace manual work? - generating jobless professionals? AI s advent into healthcare raises both risks and benefits. Careful and unbiased scrutiny of the developments can pave way for good healthcare system. Reference Jiang, F., Jiang, Y., Zhi, H., Dong, Y., Li, H., Ma, S., Wang, Y., Dong, Q., Shen, H., Wang, Y. (2017). Artificial intelligence in healthcare: past, present and future. Stroke and vascular neurology, 2(4), Keel, S., & van Wijngaarden, P. (2018). The eye in AI: artificial intelligence in ophthalmology. Clinical & experimental ophthalmology. 9