A.I. Versus M.D. – Either Way – Pathologists Do Not Diagnose Cancer – Part 1

| April 3, 2017

AIvMDA loyal reader tipped me off to a recent article in The New Yorker entitled “A.I. Versus M.D.” The article is written by Dr. Siddhartha Mukherjee, a well known oncologist and author of The Emperor of all Maladies. This article, the book and the documentary are all worth reading/viewing if you are not familiar with Dr. Mukherjee’s prose, style, professionalism and his ability to tell stories of what attracts all of us to medicine – the human spirit, discovery, curiosity, dedication and helping others.

It has been said by many more than once that “medicine is the only business continually trying to put itself out of business”.

This article in particular has many facets addressing the use of computer aided technologies, be it “artificial intelligence” or “machine learning” or “deep learning” or “deep learning automated analysis” and so forth.  Very popular buzz words right now to optimize your SEO rankings. All the rage, IBM, Google and dozens of other companies, big and small are creating warehouses of data – clinical, molecular, -omics, images, outcomes, treatments, diet behaviors, social behaviors, etc… and trying to find pieces that fit together to be personal, predictive and pre-emptive. This was the official war cry of the NIH I think more than 25 years ago – “the 3 P’s” as I recall as a resident at Bethesda Naval Hospital across Wisconsin Avenue from NIH.

Companies involved with digital pathology are increasingly focused on “analytics” and “machine learning” and automation that will allow pathologists to work more efficiently with equal or better accuracy to what we can see with our occipital lobes and try to recognize with what is between our ears whether it is taught or learned or pattern recognition or acquired through osmosis – we are trying new technologies to help us get the right answer more efficiently and perhaps with better specificity than we can do on our own.

I have many thoughts on all of this so this is Part 1 of at least 2 posts to try to stick to how much people many actually read on a blog given attention spans in 2017.

This article points out, with regards to melanoma, “If a dermatologist can do it, then a machine should be able to do it as well,” Thrun reasoned. “Perhaps a machine could do it even better.”


But one of the issues here is what constitutes a diagnosis. Is melanoma a clinical or pathologic diagnosis or both? True, hard core surgical pathologists will argue that cancer is a pathologic diagnosis. We diagnose cancer. No one else.  And depending on where one trained or who they trained with or their own convictions and those of their colleagues, some of us believe that cancer is a diagnosis by a single pathologist and perhaps others think that cancer is a diagnosis by consensus. Some cancers are diagnosed then reviewed by others for concurrence, some are concurrently diagnosed by a group at the same time, assuming there is consensus. Standard clinical business practices require quality assurance and both are reasonable approaches – again it depends on one’s practice styles formed over years.

Seldom wrong but never in doubt, as some former mentors taught me.

The reader who referred this piece to me and perhaps other pathologists going through the article will point out that the oncologist author speaks of dermatologists diagnosing melanoma and comparing what machine versus man or woman might lead to with missing the point that melanoma is a pathologic diagnosis or is it?

One academic dermatopathologist told me that about 70% of the cases that pass under her microscope confirm the clinical suspicion. A private practice dermatopathologist told me that 95% of the time or more, what he sees, the clinician has determined and he confirms the diagnosis. Perhaps academic versus community and housestaff versus seasoned dermatologists explain the difference if their numbers are right. In my experience in general surgical pathology, if it comes in a seborrheic keratosis (SK) or a basal cell carcinoma (BCC) or actinic keratosis (AK), it ends up being the same thing after all the wax and glass on the pathology report.  “R/O melanoma” usually is, meaning “ruled out” but there are occasional basal cells that are melanoma or keratosis that are squamous cell carcinoma. Which is why the biopsies are performed.

While textbooks publish images of different conditions, these are often “classic” clinical examples that may not even require a biopsy to diagnose, particularly in dermatology, but not all patients read the textbooks first to know how to present classically so they get biopsies. In transplant pathology or brain tumors, for example, a biopsy may be the only option to diagnose, grade or stage a tumor or medical condition.

But I have given up, finally, thinking that pathologists diagnose disease. Despite the dictionary definition of “a specialist in pathology; specifically: one who interprets and diagnoses the changes caused by disease in tissues and body fluids”. I think what we actually do more than diagnose disease is confirm disease, or at least confirm the clinical impression. We confirm that the SK, BCC or AK almost always is; with seasoned clinicians and pathologists, perhaps 19 times out of 20. The one in 20 are the exceptions that justify the necessity to biopsy the other 19.

Of course there are also the cases that “were thought to be X by the clinician and turned out to be Y” or “was diagnosed by radiology as Y and turned out to be Z”, proving those silly clinicians and radiologists stone cold wrong.

These cases show up the after dinner USCAP seminars until 10 PM at night. Some of those for next years meeting may have already been identified to discuss next year.

So, laymen, patients, the media, popular press and even the back of seat airline magazines tell us about famous celebrities and common folk whose primary care provider, internist, cardiologist, surgeon, radiologist or perhaps even their German shepherd [link here] diagnosed their cancer.

“Surgeons declared the patient free of cancer after removing the malignant tissue and confirming no malignant cells were present in the remaining normal tissue by looking through a microscope.” I have read this for CEOs, Popes, Senators, celebrities, even President Ronald Reagan, at Bethesda Naval Hospital, long before I ever had a parking permit there.

Of course surgeons don’t do this, at least they usually don’t drive the microscope, they might throw the microscope, but they usually won’t push the slide around, just the pathologist. But we get to leave the hospital before they do.

But for years pathologists have not really been clear on our messaging and it has impacted our specialty in many ways as I have written about previously.  When the American Osteopathic Association came out with “Image don’t diagnose patients, we DO” and “Laboratories don’t diagnose cancer, we DO”, there was a little excitement, the billboards out in front of the AOA building at the bus stop were taken down.  The others one for the expressways were never put up.

But we do a poor job of this, so poor that a well-respected oncologist points out that dermatologists diagnose melanoma and perhaps a machine could even do better than a dermatologist.  In the end, we know who told the dermatologist it was so wide by so deep with so many mitoses, with/without ulceration and so forth with largely standardized reporting we came up with, when he/she told the oncologist that he/she diagnosed a melanoma and “it came back from the lab as a T3.”

Yes, someday, robots will inhabit the earth, that man built and designed and programmed, with the ability to diagnose disease better than man, be it clinical or pathologic, before they destroy all of humanity leaving the earth scorched beyond recognition.

Hopefully, I will be retired from clinical practice by then.

In part 2 – my thoughts on man versus machine for pathology diagnosis or at least pathology confirmations.


Tags: , , , , , , ,

Category: Advocacy, Anatomic Pathology, Books, Business, Clinical Laboratories, Clinical Pathology, Current Affairs, Data Management, Digital Pathology News, Education, Electronic Medical Records, File format, General Healthcare News, Genetics, Histology, Image Analysis, Informatics, Laboratory Informatics, Laboratory Information Systems, Laboratory Management & Operations, Medical Research, Microscopy, Pathology News, Personal

Comments (3)

Trackback URL | Comments RSS Feed

  1. Rob says:

    Looking forward to part 2!

  2. Janneth says:

    La máquina no hará el diagnóstico diferencial, no se pondrá interrogantes para el diagnóstico.

  3. Hi Dr. Kaplan,

    Interesting commentary on how the “diagnosis” is made. I bet several pathologists, when the read that they don’t make the diagnosis, get all worked up. I think you make some good points though. I would give pathologists a little bit more credit for making a diagnosis but I agree we often just confirm the clinical diagnosis.

    Looking forward to part 2.

    Justin D. Richey, MD

    I’ve got a few posts in the pipeline at pathspective.com. Be on the look out. :)