Featured Post

Who are the true heroes of the future of health care?

Computers or human physicians? The answer is surprising
Illustrative photo of doctors on their way to an emergency room, October 31, 2012. (Moshe Shai/Flash90)
Illustrative photo of doctors on their way to an emergency room, October 31, 2012. (Moshe Shai/Flash90)

This is an email that I sent to a number of friends/colleagues who are working in the field of medicine and/or machine learning. I wrote the email based on an abstract that is included at the bottom of this post. I know that the language is very technical in the abstract, but the key issue is that in a specific group of patients, a “brain bleed” was missed by the initial radiologist who interpreted the CT, in 1/2 the cases reviewed.

In the early years of the state of Israel, doctors from around the world  came and contributed their knowledge to the well-being of the Israeli people. The overall level of medical technology varied between countries of origin. But the differences were relatively small for doctors from the westernized countries. During the major emigration of Russian Jews, from Russia to Israel, many were physicians. But in the 20 years preceding this emigration, westernized technology had taken a huge step forward. One perfect example was that gynecologists in this group of Russian doctors had no experience with ultrasound,  because US was considered a military device. The amazing thing about these doctors is how many of them upgraded their skills, in a language they did not initially know, and became a critical part of the health care system in Israel.

In many ways, we are experiencing the same technology gap amongst physicians today. And I am not even talking about integrating machine learning  into electronic health records. I consult to a number of startups, and I received a phone call from the CEO of one of them, who was in a bind. He had two specialists on the team of his startup, and they had literally diametrically opposed opinions on the issue at hand. The problem is that they could not point to specific literature that proved their point. More so, they both had the attitude that their experience, as a combination of their personal work and reading, with sufficient  to decide on the matter. This is not a question of lacking in knowledge or skills. This is a fundamental problem in the thought process that these doctors have. They still have not adjusted to a new reality where data decides. In ranking quality of research conclusions, personal experience is considered the weakest form. Yet far too many physicians, especially over a given age, still feel that their independent knowledge base and understanding of the literature is sufficient and unchallengeable.

When I read an abstract like the one below, it literally hurts my heart. I remember as a medical student, playing with my pet dinosaur, and asking my seniors “how do you know …. ?” And it was incredibly frustrating that, on far too often an occasion, I did not get a valid answer. As time went by, and Google was stolen from Mount Olympus and given to us by Prometheus, I began doing a great deal of reading. And I found myself having basic principles being challenged or negated on a regular basis. I also began to read about the tremendous discrepancy between medical management in different locations. In one hospital, a protocol would state one thing and in another hospital a protocol would state something contrary. I began to appreciate the tremendous amount of human error that was a part of the day-to-day work in medicine. Far too often, I witnessed radiologists arguing over the significance of a finding, or a radiologist arguing with another specialist over the significance of a finding. In the end, the patient was treated, but far from in an ideal fashion.

Human physicians are terrified of the upcoming reality whereby computers will be able to do much of what physicians do, but far better. A doctor who has just graduated medical school is totally unprepared for a reality in 20 to 30 years from now, when computer-based diagnoses will be more often correct than those of humans. No medical school that I am aware of, has a curriculum intended to train doctors to continue  studying every day,  and to be constantly upgrading their skills and adjusting to new technologies. This particular abstract below, has to do with a dramatic discrepancy between radiological interpretation of a life critical study. I think there is no question for those who are acquainted with  the capabilities of machine learning, that it will only be a machine learning system that generates a reproducible and reliable interpretation of radiological studies. The gold standard will not be the radiologist who has been practicing for 40 years. The gold standard will be [hopefully] the universally standardized computer system that reads all films.

Physicians can be very intimidating. If for no other reason, I am happy that I became a physician so that I could challenge the unfounded claims of other physicians. It is not meant to be a competition. Ideally, all physicians should be able to come to the same conclusion based on the same data. But that is simply not the case. I say to all of you that machine learning and artificial intelligence will save more lives than any group of human physicians. Yes, it is still a matter of time until these technologies sufficiently mature to overtake their human “masters”. And yes, doctors will use all means to continue to argue that human doctors are still the central and critical part of healthcare. But people like you are the ones who will cure diseases, when human doctors are at a loss. People like you will bring top quality healthcare to the entire world, including all those places where human doctors do not tread. People like you are the heroes of the future of medicine.

I’ve written before that I am taking a year to restudy my probability and calculus, before I begin a formal study of machine learning. I think my white matter is already too petrified for me to ever become truly expert in this field. But I hope that in time, that people like you will design tools that allow even a high school student to do the highest level work. And I definitely hope to be there to take advantage of these tools to study the data that is growing exponentially as we speak.

I wish you all the best and I thank you for your time in reading this

Aneurysmal Subarachnoid Hemorrhage Is Often Missed on Initial Head CT
Cara Adler, MS, Ali S. Raja, MD, MBA, MPH, FACEP  reviewing Mark DG et al. Acad Emerg Med 2016 Apr 17.

To determine how often initial noncontrast head computed tomography (CT) scans may be misinterpreted as normal in emergency department patients with aneurysmal subarachnoid hemorrhage (aSAH), investigators retrospectively reviewed records at a large health delivery system between 2007 and 2013. They identified 452 patients with a final diagnosis of aSAH, of whom 18 (4%) were diagnosed by lumbar puncture and angiography after their initial CT scans were interpreted as negative by a general radiologist or neuroradiologist.


Two independent, blinded, board-certified neuroradiologists reviewed the initial CT images for these 18 patients and rated them as showing definite (e.g., hyperattenuation in basilar cisterns), probable (e.g., isoattenuation in basilar cisterns), or no evidence of SAH. They deemed that there was definite or probable evidence of SAH in 9 patients (50%) overall and in 5 of 7 patients (71%) who underwent CT within 6 hours of headache onset. Agreement between the two neuroradiologists was 83% for definite SAH and 72% for definite or probable SAH. In the single case in which one reviewer found definite evidence and the other none, their interpretations were right tentorial subdural hematoma and meningitis.


A number of recent studies suggest that in patients presenting within 6 hours of onset of headache, a negative head CT effectively excludes aSAH. The current study’s results highlight two things: (1) This exclusion is completely dependent on radiologist skill, and it is incumbent upon ordering providers to effectively communicate potential diagnoses to radiologists so that they know what diagnoses to look for; and (2) lumbar puncture is likely still appropriate for certain patients, especially those thought to be at high risk for aSAH (NEJM JW Emerg Med Nov 2013 and JAMA 2013; 310:1248).



Mark DG et al. False-negative interpretations of cranial computed tomography in aneurysmal subarachnoid hemorrhage. Acad Emerg Med 2016 Apr 17; [e-pub]. (

About the Author
Dr. Nahum Kovalski received his bachelor's of science in computer science and his medical degree in Canada. He came to Israel in 1991 and married his wife of 22 years in 1992. He has 3 amazing children and has lived in Jerusalem since making Aliyah. Dr. Kovalski was with TEREM Emergency Medical Services for 21 years until June of 2014, and is now a private consultant on medicine and technology.
Related Topics
Related Posts