“AI combined with robotics and the Internet of Medical Things (IoMT) could potentially be the new nervous system for healthcare.”NITI Aayog
India’s think tank NITI Aayog, in its 2018 report, put healthcare on top priority in the list of domains that need an AI push. India is the perfect petri dish for enterprises and institutions globally to develop scalable solutions which can be easily implemented in the rest of the developing and emerging economies. Simply put, solve for India means to solve for 40% or more of the world. The report also anticipated AI application in healthcare can help overcome high barriers within the healthcare domain, particularly in rural areas that suffer from poor connectivity and limited supply of healthcare personnel. This is where AI-driven diagnostics, personalised treatment, early identification of potential pandemics, and imaging diagnostics come handy.
However, AI-driven diagnostics is challenging. Firstly, data availability for different age groups, genders, and regions is not adequate to make a smart machine learning model. The biases show up in the results. Underrepresented communities translate to lesser columns, which in turn translates to inaccurate diagnosis for that group.
Data is not inert, writes Rachel Thomas of fast.ai. “A software bug is not just an error in a line of code; it is a woman with cerebral palsy losing the home health aide she relies on in daily life,” warns Rachel. She also highlighted the systematic misrepresentation in medical datasets. For instance, diagnosis delays lead to incomplete and incorrect data at any one snapshot in time. Highlighting the woes of misrepresentation, Rachel said it takes five years and five doctors for patients with autoimmune diseases such as multiple sclerosis to get a diagnosis — of which three-quarters of patients are women.
Whereas, the diagnosis of Crohn’s disease takes twelve months for men and twenty months for women. This leads to incomplete and missing data. Additionally, poorly understood diseases make matters worse for any organisation trying to leverage AI in a medical setup. To tackle the most pressing challenges, Rachel recommends to focus on five principles:
- Acknowledge that medical data can be incomplete, incorrect, missing, and biased.
- Recognise how ML systems can result in centralising power at the expense of patients and health care providers alike.
- Machine learning designers must emphasise on how new systems will interface with medical systems.
- Recognise that patients have their own expertise distinct from doctors.
- A shift of focus from bias and fairness to focus on power and participation.
Rachel also insists on having a broad view of domain expertise. Though there is no doubt that the doctors with specialisations are critical to validate the models, she would like to see patients play a more strategic role in this effort. The feedback from patients will also help in building models that reflect reality in their results. “Data are not bricks to be stacked, oil to be drilled, gold to be mined, opportunities to be harvested. Data are humans to be seen, maybe loved, hopefully taken care of,” wrote Rachel quoting AI researcher Inioluwa Deborah Raji.
Lessons From The FDA
Machine learning models are already getting good at helping radiologists. But how legit are these algorithms? Have they been thoroughly vetted by government bodies? For example, when Google AI, in partnership with the Ministry of Public Health in Thailand, conducted deep learning experiments in a handful of clinics, they found fundamental issues in the way the deep learning systems were deployed. Though the model improved regularly, the challenges came from factors external to the model. Software as a medical device or SaMD comes with lots of challenges. To address such issues, last month, the United States watchdog FDA published an action plan:
- FDA’s SaMD specifications describe “what” aspects the manufacturer intends to change through learning, and their Algorithm Change Protocol (ACP) explains “how” the algorithm will learn and change while remaining safe and effective.
- FDA will make sure the Good Machine Learning Practices such as data management, feature extraction, training, interpretability, evaluation and documentation are observed.
- FDA will chart out a patient-centred approach including the need for a manufacturer’s transparency to users about the functioning of AI/ML-based devices to ensure users understand the benefits, risks, and limitations of these devices.
- To tackle bias, FDA is collaborating with institutions such as Centers for Excellence in Regulatory Science, Stanford University, and Johns Hopkins University.
- FDA will support the piloting of real-world performance monitoring by working with stakeholders on a voluntary basis.
When it comes to India, AI incorporation has to work around challenges such as shortage of qualified healthcare professionals and services and non-uniform access to healthcare across the country. India’s eHealth ambitions are yet to gain significant traction despite promising starts with initiatives such as National eHealth Authority (NeHA), Integrated Health Information Program (IHIP), and Electronic Health Record Standards for India.
According to NITI Aayog, in India, AI adoption for healthcare applications is expected to see an exponential increase in the next few years. The healthcare market globally driven by AI is expected to register an explosive CAGR of 40% through 2021, and reach $6.6 billion this year. The think tank believes the advances in technology, and interest and activity from innovators will allow India to solve some of its long-existing challenges in providing appropriate healthcare to a large section of its population. AI combined with robotics and the Internet of Medical Things (IoMT) could potentially be the new nervous system for healthcare, presenting solutions to address healthcare problems and help the government meet mission critical objectives.