Artificial intelligence in healthcare, just like AI in general, mimics neurons’ structure and human brain organization in a very simplistic but very powerful way. It approximates its conclusions without direct human input while analyzing complex medical data. The essence of AI usage in healthcare is to analyze relationships between prevention, treatment, and outcome of human illnesses. The AI in healthcare is specific since it cannot be disruptive in a way it can be in other industries. Doctors don’t want to be disrupted. They would rather adopt a tool that would ease their administrative burden so they could focus on their patients. Another set of tools, doctors would approve, are some assistance tools, that would help them with problematic differential diagnoses. Finally, tools that provide therapeutic and surgical assistance would also be in demand. As a conclusion, the process of implementing AI methodologies in the healthcare industry cannot be disruptive. It must be gradual, thoroughly tested, proven and understood. There are simply too many moral, ethical and legal implications for doctors to just embrace novelty in a way that is advertised and embraced in other industries.

Eager to learn how to build Deep Learning systems using Tensorflow 2 and Python? Get the ebook here!

In order for something to be fully implemented in healthcare, its mechanisms have to be fully and precisely understood since an error can lead to health impairment or even death. Hence, the health legislature is very rigid. In addition, some of the AI methods are still “black boxes” meaning that nobody knows the exact logic behind AI’s predictions and why a certain prediction has been made. That causes a lack of confidence in the technology by both doctors and patients and could potentially cause another AI winter. Fortunately, there are a couple of good reasons for us to believe that this time the AI winter is not coming – the abundance of health data (from heart rate to genotype) which can be combined with data from other trackers (social media, GPS, billing data, etc.), and the immense computing power, especially in cloud services give us indications that AI will be a hot topic in the field for a while.

Until now, a huge amount of medical data stored in hospitals worldwide was fragmented, hard to collect and difficult to analyze. Now, for the first time in history, we have the ability to collect and analyze all those zettabytes of information which are continuously generated in the healthcare industry (that is a zettabyte per year). And this is a lot indeed since one zettabyte is a trillion gigabytes. Thanks to this information, it has become possible to analyze a human being in multiple dimensions (so to say) – biological, environmental and social. AI algorithms are finding all kinds of patterns in data which is comprised of static data such as EHR, diagnostics and genetic analyses as well as dynamic data such as variable sensors and monitors, and also the data acquired from social media. This will impact physicians, health systems and patients. This approach will assist physicians in diagnostics and therapy, both conservative by tailoring patient-specific therapeutics and surgical such as AR, MR, and robotics. It will also assist health systems in improving efficiency and cost reductions. Finally, it will increase patients’ overall satisfaction by improving their experience in contact with the healthcare system as well as by improving the outcome of treatments.

Fields of Artificial Intelligence Application

There are three main areas of AI application in healthcare: 

  1. Medicine itself – prevention including gene editing, diagnostics, efficient drug development, and therapy.
  2. Assistance – monitoring (wearable sensors, ambient intelligence – hospital and home continuous video capture) and telehealth (digital consultants, chatbots).
  3. Administrative tasks – EHR, voice to speech technologies, etc.

Medical Specialties and Artificial Intelligence

To say, this list is by no means finite. The AI research and development in healthcare are growing rapidly and updates are coming on a weekly and daily basis so no matter how fast a paper is published there is a good chance for it to be outdated in a matter of weeks or days if not immediately after publishing. So, here is the list of the current AI implementations in healthcare:


A medical field currently most impacted by AI technologies which are at their strongest in image recognition. These techniques are successfully implemented in prevention and diagnostics protocols. These include fracture detection, soft tissue injuries and disorders, and tumors. 


Used for detection and classification of tumors as well as for disease outcome predictions – immunotherapy and chemotherapy response, radiotherapy toxicity and survival prediction using imaging, genomic and EHR data.


AI methods are implemented in laboratory analyses, differential diagnoses for diseases of red and white blood cell lineage. Diagnosing malignancies based on the monitoring of gene expressions – moving the traditional diagnostic pathways from clinical to molecular-based diagnostic systems.


Used for non-invasive diagnostics such as ECG, nuclear radiology, cardiac CT and MRI scans as well as echocardiography.


Currently, the applications refer to Gleason score prediction by recognizing cancer cells from digital pathology slides of the prostate biopsy specimen, and Tumor proliferation score of breast malignancy in the breast resection specimen.


AI methods are used for the assessment of spermatozoids, ovarian reserve parameters, and embryos quality.


Used for diagnostics of malignant skin diseases (melanoma, basal cell carcinoma, and squamous cell carcinoma) and chronic skin diseases (psoriasis), all based on image recognition. 

Orthopedic surgery

There are currently developed algorithms for fractures and malignancies detection based on X-ray, CT and MRI image recognition. Also, there are a number of commercialized AI algorithms for predicting injury patterns and predicting postoperative complications following orthopedic and trauma procedures.


This is one of the medical fields with the most abundant AI implementation. It includes prediction of child brain maturity, prediction of psychosis and unipolar depression, evaluating attention-deficit/hyperactivity disorder, evaluating risk for anxiety, seizure prediction in children with epilepsy, identifying motor abnormalities, appendicitis risk stratification in ER, detection of low-volume blood loss, identification of regenerating bone marrow cell population, gene expression profiling for children with lymphoblastic leukemia.


AI algorithms are being used for suicide prediction and for depression and anxiety treatment, a feature performed by chatbots.


Used for early diagnostics of chronic diseases such as Multiple sclerosis, Alzheimer’s disease, and Parkinson’s disease, and for a number of acute neurological diseases such as brain tissue ischemia, intracranial hemorrhage, and hydrocephalus.


Image detection of age-related macular degeneration, macular edema and retinopathy represent some of the most developed AI algorithms in healthcare.


Used for chest X-rays image detection of malignancies, for pneumonia diagnostics and classification, for assistance in pulmonary auscultation and interpretation of pulmonary function tests.


Having said all of the above, we cannot miss mentioning the hype surrounding the AI evolution in medicine. Although powerful, it is not magic and it comes with its associated risks. One of the most prominent risks is social and experience-based bias, which humans inadvertently transfer to AI algorithms influencing the final result. This is further complicated by the “black box” problem making it more difficult to comprehend the underlying mechanism. Then there is a trust issue. Since the real clinical practice data is often messy or incomplete, an AI system working on that data can’t be 100% accurate so humans must know when to trust it and when not to trust it. Apart from that, another thing we should take into consideration is potential legal issues. Who is responsible if something goes wrong? A doctor who is using the AI method or a programmer who developed it? 

At this point, the main technical problem with a wide implementation of AI into healthcare is that the full automation in medicine is hard to accomplish. AI is superior in rigid decision-making but in a dynamic environment such as health, there are many changing variables and incomplete data. That level of automation is much harder to accomplish. All of the aforementioned points to the fact that AI disruption of health is not possible. Instead, this continuing process we are all witnessing would be properly named the AI transformation of health.


AI algorithms are being implemented in diagnostics, treatment protocol developments, drug development, and personalization as well as short and long term personal and ambient monitoring making AI an integral part of healthcare by enhancing it to be more efficient, accurate, personalized and cost-effective. Much more of it is yet to come since there is a great number of research projects currently going on in the AI field regarding the healthcare industry but that is outside the scope of this article. 

In the end, the whole purpose of the AI is to improve treatment outcomes to the satisfaction of both patients and doctors which is the ultimate goal of medicine itself. 

Nemanja Kovacev

Nemanja Kovacev

MD, Ph.D., Orthopedic and Trauma Surgeon, Java and Python developer, Health Tech SME

Nemanja Kovačev is a health tech subject-matter expert with 15 years of combined medical and programming experience. He defended his ph.d. thesis at the University of Novi Sad, Serbia. Participated in national and international medical and IT conferences as a lecturer and a session chair. Loves to write scientific and popular science articles. Currently a full-time programmer and a health tech SME.