Since the past few years, the applications of AI have shown incredible promise in a wide variety of industries. This enthusiasm has also grown amongst healthcare professionals and its capabilities have always been portrayed positively and its potential benefits stretched further than where they are.
As AI has become increasingly mainstream, now is probably a good time to assess the areas in which the industry has benefited from this disruptive technology and also, perhaps critically, gotten ahead of itself when it comes to the technology’s capabilities and applications.
Let’s look at three assumptions about Machine Learning technology in healthcare that are considered to be misconceptions:
1 – That machine learning tool can do much of what doctors do.
At least for the foreseeable future, machine learning applications will not be able to replicate most of a doctor’s capabilities. Although, currently, machine learning applications can help doctors in prevention against illnesses and diagnosing medical problems, especially when it concerns analyzing images to detect any complications. One thing ML applications will not be able to do is provide the care and unique treatment according to a patient’s condition anytime soon.
Although ML applications are capable of running diagnostics, they still require their output to be analyzed by someone with domain knowledge, as there is the possibility for trivial data to be interpreted as essential. Another aspect that is lacking in ML applications is the ability to play a role in helping patients decide whether to receive treatment and the kind of treatment that would be suited to their lifestyle and capabilities.
2 – That “big data” + brilliant data scientists are always a recipe for success.
It is true that experienced and skilled data scientists are crucial for developing sophisticated ML models, but that is not the end of it. Domain experts are also equally as important as they understand the requirements of the model and how to formulate it to receive the desired outputs from large databases.
One method of ensuring success is to implement an ML-plus-human approach. In this method, for example, the ML algorithm makes its decision regarding categorizing medical supplies and drugs after which a data scientist can review the output provided by the ML. This helps to smooth out the process and fix problems and inaccuracies, especially those which fall in a lower confidence range.
3 – That healthcare leaders will adopt and use successful algorithms once they have been discovered.
A lot of promising and powerful algorithms do not get widely utilized and adopted due to them not being integrated into the workflow of potential users. Inaccurate adoption sometimes leads to wastage of crucial time for doctors as they have to go out of their usual processes to access and activate the ML applications. As a result, these ML applications end up being used and neglected.
But when they are integrated correctly, ML applications have shown tremendous success for automatically scanning incoming faxes, papers and other electronically accessible documents relating to consent for surgery and filling them in the correct medical record. Such ML applications are also capable of having an alert for the pre-operative checklist. This has shown incredible promise as it can save up to 120 hours of time per month for medical staff.
Such myths are misconceptions like this technology can magically solve all medical problems or replace medical personnel who are present whenever there is any new and emerging technology. Although the applications have shown that the correct implementation and integration into the workflow of medical facilities will lead to a better system for operations which have lower frequencies of errors. They are also critical in saving time for medical staff as they can deal with repetitive and mundane tasks without error and fatigue.