Techloging – Tech Software/Hardware, How-tos, Reviews and Guides Feel the magic of tech
Related Articles
The Reality of Artificial Intelligence in Health
It is not possible to think about the future of health care without having two entiries which, together, represent the hopes and fears of an industry that is looking for smarter solutions. We’re talking – how could it be otherwise – of artificial intelligence.
Artificial Intelligence (AI) has existed since 1956, and in the 1970s it had its first experience in the health sector with the so-called Mycin, an expert system aimed at detecting infectious blood diseases that reasoned, communicated in natural language with the user, and prescribed medication in a personalized way for each patient.
If you were already talking about intelligence, Natural Language Processing (NLP) so early, you may wonder why there are so few contributions of AI to medical practice and why it seems that it is now when learning based on machines and data begins to merge with reality.
AI in medicine: types of computational intelligence
The confusion surrounding Artificial Intelligence, and even its definition, remains widespread. Today, AI is any task that a computer can perform, if not better than humans. But when we think about the application of AI in medicine, we must consider different types of computational intelligence.
Most of the machine-generated solutions emerging in healthcare today are not based solely on the intelligence of algorithms over data. Rather, the algorithms created by the experts are just the basis for analyzing data and recommending treatments.
The so-called “machine learning” is based on algorithms known as neural networks, that is, a computer system modeled after the human brain.
These applications include multilevel probabilistic analysis, allowing computers to simulate and even extend the way the human mind processes data.
As a result, even the programmers cannot be sure how their computer programs will arrive at solutions, and therefore the doctors will not know how they have arrived at an optimal recommendation.
In another variant of Artificial Intelligence, known as “deep learning“, the software learns to recognize patterns in different layers.
This mechanism is increasingly useful in medical practice, as each layer of the neural network works independently but in a coordinated way, separating aspects such as color, size, and shape before integrating the results.
These new visual tools promise to transform medical diagnosis and can even search for cancer at the level of individual cells.
The use of Artificial Intelligence in healthcare
As we see, Artificial Intelligence can be applied in many different ways, but the best way to understand its potential use in healthcare is to divide its applications into three categories:
- Algorithmic solutions
- Image processing
- Support tools for medical practice.
1. Algorithmic solutions for healthcare
In current medical practice, the most widely used applications of AI are algorithmic ones: evidence-based approaches, programmed by researchers and clinicians.
When humans integrate known data into algorithms, computers can extract information and apply it to a problem.
Let us take as an example the evolution of Multiple Sclerosis. Using algorithms in consensus with experts in the field, together with existing data in the computerized clinical history, numerous treatment alternatives can be reviewed and the most appropriate combination for a specific patient can be recommended.
Sepsis is the medical condition with the highest mortality – 13 times more than traffic accidents – and use of resources and costs, between $14,000 and $28,000 for each case of sepsis detected. To improve its detection and treatment, the development of AI and ML, is being carried out in two phases by hospitals:
- A first phase for the detection of sepsis in real time,
- And a second phase will provide clinicians with decision-making assistance, with a comprehensive approach, from diagnosis to therapeutic support for patients with suspected sepsis.
Early detection of sepsis is essential since it is a time-dependent clinical process, such as myocardial infarction or stroke. The sooner it is identified, the sooner it can be treated, with the consequent improvement in results: lower mortality, use of resources, and costs.
2. Image processing
To appreciate the potential of visual pattern recognition in medical practice, we must understand how often the human eye fails even the best clinicians.
Independent studies say that between 50% and 63% of women in the US who have had regular mammograms for 10 years will receive at least one “false positive.” That is, the test result falsely indicates the possibility of cancer, thus requiring additional tests and sometimes unnecessary procedures.
Also, up to a third of the time, two or more radiologists examining the same mammogram will disagree on the interpretation of the results.
In such studies, visual pattern recognition software, which can store and compare tens of thousands of images using the same heuristic techniques as humans, is estimated to be 5% to 10% more accurate than the average doctor. And the gap in accuracy between the human and digital eye is expected to widen further before long.
As machines become more powerful and deep learning approaches gain traction, they will continue to advance diagnostic fields such as radiology (CT, MRI, and mammography), pathology (cytology and microscopic diagnoses), dermatology (rash identification and evaluation of pigmented lesions for potential melanoma) and ophthalmology (examination of the retinal vessels to predict the risk of diabetic retinopathy and cardiovascular disease). And these are just a few examples.
3. Support tools for medical practice
In the television series House, a doctor’s genius trumps the expertise of his colleagues. We might think that if all doctors were as smart as Dr. Gregory House, the puzzles associated with diagnosis would disappear.
Experts suggest that the biggest difference between doctors is not their level of intelligence, but: (a) how they approach patients’ problems and (b) the health systems that support them.
Because “a” and “b” combine to create wide variations in clinical outcomes, Artificial Intelligence holds great promise for the future. Two AI approaches in particular, both currently available, could radically improve physician performance.
The first is Natural Language Processing, a branch of AI that helps machines understand and interpret human speech and writing. This software can review thousands of complete electronic medical records and figure out the best steps to take to assess and treat patients with different illnesses.
The second approach involves the use of computers to observe and learn from doctors as they go about their business.
Instead of retrospectively extracting and analyzing data, AI and machine learning follow what clinicians do step by step. With the incorporation by the professional of the patient’s data into the digital medical record, so many AI tools analyze all this information in real-time to offer an early warning and suggest possible personalized treatment. In this case, false positives are halved and their effectiveness is maximum.
Unfortunately, the biggest barrier to AI in medicine isn’t math. Rather, it is a medical culture that places greater weight on the physician’s intuition than on evidence-based solutions provided by information and analysis.
Therefore, doctors are expected to understand that AI is not here to replace them, but that those professionals who use AI will replace those who do not.
The potential of AI in health
Artificial Intelligence offers a potential that goes beyond that shown in this article. We find an example in the experience with an ancient Chinese game invented more than 2,500 years ago: «Go». In this two-player board game, opponents try to claim the most territory in an incredibly complex and abstract game, with a seemingly infinite set of possible moves.
Go’s degree of difficulty left few observers believing that a machine could be better than a competent human. A challenge was achieved in 2015, when AlphaGo, a program created by the Google Deepmind division, beat Lee Se-dol, one of the best Go players in the world.
The most interesting thing, however, is how he did it. Unlike IBM’s Deep Blue, which defeated chess champion Garry Kasparov in 1997, AlphaGo did not “learn” by studying humans and repeating previous games.
According to an article in Nature, humans may have taught AlphaGo the rules of Go, but it is the program itself that has mastered the game by playing against itself.
This type of “deep learning” could be the very thing that drives the application of AI in health in the future, as it would help to obtain better medical care, create new approaches to diagnose and treat hundreds of medical problems, and measure the adherence of doctors without the faulty biases of the human mind.
Additionally, and over time, patients will be able to use a variety of AI-based tools to take care of themselves, just as they manage many other aspects of their lives today.
