Li Feifei's team Nature: How to illuminate the "dark space" in the medical field?

In their paper, the authors of "Environmental Intelligence" and "Dark Spaces" point out that the development of machine learning and contactless sensors has led to the development of "environmental intelligence".
refers to a technique that produces a sensitive response to human existence in physical space.
machine learning, many people may think about artificial intelligence and its application in disease diagnosis.
, artificial intelligence-assisted decision-making has become more common in some medical settings.
, but how these decisions translate into specific physical behavior by doctors, patients, or their families remains unclear.
of key medical practices in hospitals and at home remains undefined.
, the researchers likened what we don't yet know to a "dark space."
it's not just a metaphor.
up to 400,000 people die each year in the United States from medical decisions or in actual negligence and error, according to the World Health Organization (WHO) and the National Institutes of Health (NIH).
the development of artificial intelligence has the potential to assist doctors, not only in assisted medical decision-making, but also in improving specific medical steps. The
section can be used for "environmental intelligence" sensors (Photo Source: Resources) while passively sensing sensors installed in the environment can act as "environmental intelligence" to understand people's movements and their medical needs, thus better helping doctors and service providers implement health initiatives to the ground.
hospitals are used in hospitals around the world, where they are the primary places for medical behavior.
, 7.4 per cent of the US population will stay in hospital for one night in 2018, up from 17m in the UK.
, environmental intelligence is expected to play an important role in the hospital environment.
the first application mentioned by the researchers was the Intensive Care Unit (ICU) for potential applications in hospitals for "environmental intelligence" (Photo source: Resources1).
application is to monitor the patient's mobility: for patients with critical conditions, neuromuscular weakness often occurs, and allowing the patient to regain mobility as early as possible can reduce related debilitating events by 40 percent.
For now, the standard method of evaluation is artificial direct observation, but this introduces bias and error, and the cost is uncontrollable, while wearables can only detect dynamic switching and cannot take the environment into account (e.g., sitting in a chair or on a bed).
environmental intelligence can avoid these problems.
in one case, machine learning algorithms were able to distinguish between whether a patient was lying in bed and walking with 87 percent accuracy.
another application in the ICU is to control infections - 30 percent of ICU patients develop hospital-related infections each year, and hand washing is one of the most effective ways to reduce infections.
environmental intelligence can efficiently assess whether hand washing is compliant and does not require people to count.
a deep learning algorithm that assessed 351 hand washing incidents in an hour with a 75% accuracy.
the predictive accuracy and human statistics of "environmental intelligence" are very close (Photo source: Reference 1) ICU, the researchers point out that environmental intelligence can also be used in operating rooms.
, for example, can assess a physician's level of surgery based on the trajectory of the needle driver during surgery.
the algorithm is 92% accurate compared to human assessment.
other algorithms can take an inventory of the tools used in surgery and avoid missing them in patients.
this adds a layer of double insurance to the doctors' daily inventory.
the future, video-based surgical phase recognition is also expected to better train surgeons.
, the authors of the paper also point out that environmental intelligence can help doctors with text-recording work, which doctors spend about 35 percent of their time on.
the microphones installed in the environment, the machine automatically recognizes conversations between doctors and patients with more accuracy than manual listening.
, it compresses two hours of text finishing to 15 minutes, allowing doctors to spend more time with patients.
applications in everyday life In the ageing world, an application of environmental intelligence is the place where older people live.
by 2050, when 1.5 billion people over the age of 65 worldwide learn about their daily lives, it will be key to taking care of their health.
, the day-to-day living behaviour of older persons is mainly through self-reporting or caregivers' points.
some studies have found that environmental intelligence can accurately assess when older people need help, with an 86 percent accuracy rate.
, bathe, or go to the bathroom, you can replace the camera with a microphone for privacy.
another application for older people is to monitor for falls.
, environmental intelligence is as accurate as wearables, at 97% and 96%, respectively.
important for early detection of elderly falls and early intervention.
potential scenarios for "environmental intelligence" in everyday life (Photo Source: Resources) In everyday life, the second application scenario for environmental intelligence is chronic disease management.
is important for children with cerebral palsy, or for people with Parkinson's disease.
, this has traditionally required specialized laboratory evaluation, and wearables are prone to patient discomfort and errors.
contactless environmental intelligence improves credibility and allows you to schedule a rehabilitation program at home.
paper, the authors point out that the third application in everyday life is mental illness, such as depression, anxiety disorder, or bidirectional personality disorder.
, the diagnosis of these diseases is based on self-reporting and medical scale assessment.
and environmental intelligence can continuously monitor a patient's symptoms.
study found that by analyzing data such as speech and video, schizophrenia was detected from speech and upper body movements, with a sensitivity of 84 percent.
, environmental intelligence can also identify suicidal tendencies.
challenges and other considerations, researchers also point out that for environmental science to be more widely applied, we need to address algorithmic challenges and understand human behavior in complex places.
we can learn a lot from big data.
in practice, privacy is the number one issue we need to consider.
, environmental intelligence continuously monitors information and human behavior in the environment, which inevitably introduces privacy concerns.
privacy is a must when developing technologies for medical environments.
, some practices prevent the disclosure of personal privacy information.
addition, we need to avoid unauthorized access and further eliminate the possibility of disclosure.
some technologies can protect patients' privacy (Photo Source: Resources) Finally, there are potential risks to environmental intelligence, such as fairness, transparency, and research ethics.
can we get different people to benefit from technology? How do you get interpretable data from environmental intelligence instead of the usual black box of information? How to comply with research ethics and maximize the protection of individuals involved in research? These questions still need to be answered.
conclusion of the paper, Professor Li Feifei's team concluded that breakthroughs in artificial intelligence technology and the advent of low-cost sensors have made environmental intelligence a potentially important tool for optimizing medical decision-making and execution.
, some research data have explored the feasibility of this technology.
future, we also look forward to this technology can illuminate the dark space of medicine for the benefit of human health! References: s1. Haque, A., Milstein, A. and Fei-Fei, L. Illuminating the dark spaces of healthcare with ambient intelligence. Nature 585, 193-202 (2020).