Artificial Intelligence-The stethoscope of the 21st Century
Twitter: @swatisubodh
The challenge of fewer doctors, rising healthcare costs, emerging diseases and antibiotic resistance of microbes deters scalable and workable healthcare solutions. In India there are 0.2 doctors for every 1000 Indian, this ratio is amongst the lowest in the world. Medical facilities are concentrated in urban areas with a majority of the world’s populations being left without access to even primary care. Then, when we talk about something like Artificial Intelligence (AI), doesn’t it sound out of context? Should the focus not be on bringing primary care to the door step, or increase the number of medical professionals, or further increase in number of medical colleges etc.? There is no doubt that these issues are important, and they have been in focus over many decades, however this linear thinking has not resulted in the impact that a rapidly growing population with evolving healthcare challenges needs. On the other hand, assuming artificial intelligence in healthcare has nothing to do with it will be incorrect.
AI has the potential to reshape India’s crumbling healthcare-not just for the urban elite but also for those at the bottom of the pyramid. AI has been projected to be the defining factor across many sectors; however we zoom in here on the immense potential it holds in healthcare- right from fitness tracking to split second decisions taken in an emergency room; AI is expanding in scope and stealth across the entire healthcare value chain
Highlighting the key areas where AI can impact healthcare, Dr Anurag Agarwal, Director, CSIR Institute of Genomics & Integrative Biology, New Delhi, and part of the 18 member task force to recommend on AI based interventions constituted by the Indian central government, states that routine tasks where well defined protocols exists, e.g. filling of templates, assistance in ensuring that guidelines are being met etc., are likely to be automated. He further explains that user-defined or machine-learning based monitoring of health data with early warning based on pattern recognition is another area where AI can be immensely useful. Also, pre-diagnosis in medicine and definitive reporting in image based problems, such as those used in radiology and pathology, will also see AI based interventions in a big way.
A report by PwC recently estimated the AI impact factor in healthcare on a scale of 1–5 (lowest to highest). Prime focus is foreseen in personalized approach to treatment where deviation in data from baseline levels can be identified leading to early diagnosis and intervention thereby preventing disease aggravation. Here the patient’s previous data would be used in defining baseline levels. This is also projected to save time by enabling patients to seek timely help while being able to manage their insurance and scheduling their tasks well ahead of time.
Chatbots, similar to Siri on iOS, are likely to be the first line health assistants providing the first interface in the comforts of our home. This spells huge potential in geriatric and pediatric care in the cities where it can guide an anxious patient or a care giver through basic queries and help them in reaching out to paramedics or doctor. Chatbots could also enable prescription management, reviewing symptoms & advice, reporting symptoms & illness etc. Data supported chatbots could also provide the first line of health assistance in underserved regions with poor healthcare delivery infrastructure and help people navigate through common health queries.
Another area that would benefit from AI in healthcare is drug discovery. The task of developing a new drug conventionally takes years to fructify and many billions of dollars in investment. This sector is expected to benefit from the quick identification of drug targets and prediction of structures of potential drug molecules due to AI engagement throughout this pipeline.
Artificial ‘Narrow’ Intelligence (ANI)
Although the overall sentiment of AI based interventions in healthcare is upbeat; one also has to consider that there are still limitations that we need to overcome. Although these maybe issues pertaining to the early days of AI and its application on ground which are likely to be resolved as more data sets become available and better algorithms support the backend; this still needs a mention and acknowledgement in the current scenario.
Few have termed it as Artificial ‘Narrow’ Intelligence due to its current capacity to do marginally better than manual capacities in specific tasks, eg out perform in chess but unable to do other tasks like driving a car or paint as the chess player on the other end might also be capable of doing. This Artificial ‘general’ intelligence (AGI), which is desired, is the level of intelligence when a machine becomes capable of abstracting concepts from limited experience and transferring knowledge between domains.
AI limitations, specifically in the healthcare domain, currently are;
- Image recognition, using machine learning and deep learning algorithms for the purposes of radiology runs the risk of being given thousands of images which might include underlying biases as well. This bias, which maybe previous case related or that of the working team, would have limited use in identification of novel cases.
- Streamlining & standardizing past and existing medical records, which might be incomprehensible, incomplete or fragmented might pose a problem in developing efficient algorithms
- Ethical and legal aspects pertaining to AI based diagnosis are still blur. What if AI does not identify a cancerous node in the lung and marks the case as a false negative? Or what if AI incorrectly diagnoses someone with a disease (a false positive) which leads to an onset of lengthy and expensive treatment protocols?
- Misconceptions & hype are not technical roadblocks but roadblocks nonetheless. Misconstrued information and ideas on use of AI in healthcare is preventing many from keeping an open mind and exploring this tool to its potential
It is expected that in coming 5–10 years AI based medical interventions will be out in the market and very commonplace. This is seen from the upswing of AI based startups who have already taken lead. 23% of the AI-only startups in India are focused on healthcare. Global giants like Google are bringing their AI capabilities to India as well. Google recently partnered with Aravind Eye Hospitals to use image recognition algorithms to identify retinopathy among patients with diabetes. The hospital has provided close to 128,000 retinal images to Google that have been crucial for it in developing the application of AI to detect diabetic retinopathy in 415 million at-risk diabetic patients worldwide.
Indian Healthcare.AI Startups
Diagnostics
SigTuple started its activities in 2015 after angel funding from well know angel investors. Later in raised $5.8 million from Accel Partners in series A funding. With it multiple platforms, like Manthana and Shonit, SigTuple is automating diagnostic tests from blood, urine, semen samples, retinal scans and chest x-rays. This is being used in detect diseases like anemia, malaria, leukemia and other diseases. Not only is it faster; it is more efficient and cost-effective than conventional ways of pathological or manual testing.
Aindra this 5 year old startup is funded and incubated by Villgro. It uses AI to analyze pap smear samples from women to detect cervical cancers using its in-built digital microscope with an onboard computing unit. The tests are verified by pathologists, which further enables atunement and learning of the AI platform. This will enable better detection in India which has one third of the global burden of cervical cancer cases. Through its early detection platform, Aindra can measurably increase the odds of survival. They aim to reach out to 330 million Indian women in the at-risk age bracket through their portable point-of-care platform.
Niramai Health Analytix is just over a year old and has raised seed funding from Pi Ventures, Ankur Capital, Axilor Ventures for its portable AI based portable thermal imaging device for breast examination for screening early stage breast cancer.
Qure.Ai through its AI imaging analytics aims for better diagnosis of radiology imaging based diagnosis. The deep learning algorithm makes treatment plans based on radiological & pathological image analysis for personalized cancer care.
Monitoring & Tracking
Ten3T, through its AI based palm-sized cardiac care monitor, Cicer, is making tracking of ECG, respiration, pulse, temperature easier and on real-time basis. The data is streamed to the doctor or at the clinic for monitoring the patient. It is adapted for various clinical, nursing and in-home applications.
Touchkin is a predictive healthcare app which records parameters as sleep, activity, and patterns of communication through sensors and smartphone to identify changes in behavioural pattern.
Genomics
Orbuculum has brought together AI with genomics to predict cancer, diabetes, neurological disorders, cardiovascular diseases in a fast and cost-effective way. It extracts meaningful information from the enormous amount of genomic data generated globally and utilizes it to benefit the society. The data would also be used to understand the genetic basis of many life-threatening diseases.
Precision Medicine
Moving from ‘one size fits all’ to a more ‘personalized’ approach to treatment is what precision medicine aims for. This is being made possible by person-specific data, like personal genomes. To this, add nutritional and environmental factors you one is likely to sit on huge amount of data. AI has the potential of bringing these together to be able to make better medical recommendations and predictions, not just in disease but also in assessing the best treatment regime to follow based on individual drug susceptibility profiles.
Internationally, several companies have already started harnessing A.I. for mining medical records (Google Deepmind and IBM Watson), for identifying therapies (Zephyr Health), supporting radiology (Enlitic, Arterys, 3Scan) or genomics (Deep Genomics). Atomwise uses supercomputers that root out therapies from a database of molecular structures recently identified, through its AI platform, two drugs which could potentially be safe and efficient against Ebola. Something like this would conventionally take months or years to accomplish.
Going forward
Although many ambitious projections of AI replacing the doctor behind the desk are doing rounds, evoking elation and excitement on one end; and skepticism and doom on the other; realistically speaking AI is initially likely to be adopted as an aid, rather than as replacement, for human physicians. It is projected to enhance and augment the physicians’ diagnoses and at the same time, process and provide valuable insights for the condition. This way AI will learn continuously, improve and become sharper in its future assessments which will enable AI-powered diagnostics to gain in accuracy and ultimately provide confidence in undertaking bigger and more complicated tasks. Whether this will translate to delegation and complete autonomy in functionality of certain tasks is anybody’s guess!
What is certain, however, is that in years to come the healthcare sector will use more of AI based technologies to improve not just its own efficiency but also for better healthcare delivery and research with medical professionals playing a key role in re-shaping the future of medicine.
(The article was published as part of the writer’s monthly column, ‘Kick Start’, in January 2018 issue of Nano Digest)
