Introduction

How good are we doctors at our craft? Can machines beat us in providing better patient care? Surely Medicine is too complex and every patient is different. There is no way an algorithm driven self learning software can do any good for a patient. Or can it?

source

limits of current methodology in medical research.

Often in medicine, we research treatments or interventions by comparing them to one another. We often choose large number of patients that when divided in 2 groups may look similar to each other. We need large number of patients in hopes of getting valid results. Then we treat both groups differently and tease out what works and what doesn’t work. For example look at the following famous and excellent study that impacted how we practice now.

source
Simply stated one group of very ill patients got nutrition early intravenously and the other group received nutrition started several days later. The results might surprise you but the point I am making is that the application of this knowledge is applied to the general patient population and that is pretty much the best we can do right now. Imagine how much time and data collection is needed to simply refine this just a bit more. Is feeding 2 days into the illness better than starting at 4 days. Is it possible that withholding nutrition for 5 days vs 7 days is much superior? Is it possible that these timelines should be individualized? We are about to find out the answer to these and similar questions sooner than we expect.

Is human capacity limited in the practice of medicine?

Perhaps an impossible question to answer as there are many aspects of skills in varied fields. The correct question may simply be: “how do we get better?’ Let’s look at one example. It is quite routine nowadays to get something called a CT angiogram to look for blood clots that may be present in the lungs. This is termed pulmonary embolism. A dye is pushed into circulation. A very fast CAT scanner rapidly takes images as the dye circulates through the lung blood vessels. The pictures are analyzed by a radiologist to determine whether there is a clot present. A study looked at how well we do with correct diagnosis:

source

Not surprisingly there were many scans read as positive for pulmonary embolism when none was present. This is an expected finding for reasons that are not relevant here.

What is the rate of cancer misdiagnosis?

One study reported the number to be as high as 28%.

The field of Medicine has been ever improving.

From disease eradication such as small pox, to curing certain cancers and illness such as hepatitis C, advances in medicine have been amazing. Although, many times when we have looked back we have noted the room for improvement.

ENTER ARTIFICIAL INTELLIGENCE.

Prediction of disease.

source
Prediction of disease was studied utilizing AI program named Deep Patient. Results published showed improved prediction of diseases like schizophrenia, cancer and diabetes.

---

AI reading Imaging studies.

source
This AI program was only fed 1007 X-rays of patients with TB and without TB. Even at this minute learning scale, the program was able to diagnose TB on subsequent X-rays with a 96% accuracy.

---

AI detecting metastic cancer on microscopic specimens.

source
For now the program was found to be as good or better than Pathologists.

---

The Future.

Artificial intelligence will not only improve every aspect of Medicine, it will change everything we know. Searching and learning through millions of records, it will be able to see the minutest individual variations and give us predictive models and suggest the best individualized course of therapy. It will see things differently and analyze from a prespective which a human mind cannot achieve.
We will get answers to things for which we did not know a question even existed. I expect a lot of surprises. I do not expect to practice a decade from now in any shape or form that currently constitutes the practice of Medicine. The evolution is here.

source