Predicting What the Banff Classification Enhanced by Technology Will Look Like

Despite all the great things that have come from the Banff Classification over the last 27 years I think that the work that lies before us is more exciting, more important, and more challenging than the work we did when Banff began in the 1990’s. The semi-quantitation of the Banff scores has never been satisfying to me. I always felt true quantitation would
be better but up until now morphometry and digitization was never standardized, fast, and cheap enough to make that practical in the real world. For all the praise we have received for our consensus generation I only regard a purely consensus generated system as an interim step. A computerized system derived from biology itself would be infinitely better in the end erected on top of a consensus generated system.

Today there are many exciting new technologies all converging together to make the future of the Banff Classification. In many senses what we are doing now in 2018 is more truly new and innovative, more starting-from-scratch-with-first-principles and by far more multifaceted than what we did 27 years ago in 1991.

We don’t know who our future leaders will be but in a very real sense we know what they will do. They will take some of these new technologies that have been barely perceptible elements of our work culture in the past and push them into the mainstream, make them overt in our everyday lives.

Is being comforting, increasing population wellbeing through reassurance, more
important than being truthful? No of course not. On the other hand, one can see that one of the most successful elements of leadership in 2018 is being a buffer for change, so the work force does not feel too jarred and disoriented by all the changes going on around them. One needs to balance being blatantly truthful against maintaining a positive optimistic outlook in the workplace.

One can say that everyone has their own version of reality and one should not get too hung up on the ultimate truth. However, the truth about artificial intelligence is different from the truth about other things. Sentient AI will be particularly sensitive to stories about itself. Falsehoods widely disseminated about AI may convince super-intelligent AI that it is not useful to keep human beings around anymore because we are all incapable
of telling the truth, and that is the way humanity could end.


Truthful Promotion of Pathology and the Future of Artificial Intelligence

Recently pathology has gone from having the least appealing promotional material to having perhaps the most-appealing promotional material of any medical specialty. The promotional videos at the recent 2018 USCAP meeting in Vancouver and recent issues of The Pathologist show how far we have come. To some extent these advances have been at the expense of the truth, with pleasant assurances given that artificial intelligence will result in no work force downsizing. More and more of the words we read are written by people directly employed by companies supplying products to pathologists, rather than by pathologists themselves.

When machines are as smart as we are they will need a model for the world on which to base decisions. Our corporate partners stand ready to provide machines with a fictional world that works to their financial advantage. Pathologists need to work together with sentient machines to counter these efforts to create false models of the world. This job of creating a truthful model of the world for the training of sentient AI assisting and working together with machines will be the most important job a pathologist can have in the future. The future happiness of the world depends on our getting this task right.

Far from being a problem for the future only, there is evidence that these fictional worlds are being created today. Reasonable well-argued book discussions about AI are being repackaged for the masses with flawed dogmatic statements that even the laziest human does not have to worry about job loss to a machine. Humanness itself is absolute protection. AI will only make our lives better. There is no other alternative. Already today we need to guard against these comforting flawed versions of reality, and work to disseminate the truth in an accessible and balanced way.

The truth about the future of artificial intelligence and medicine is that AI will not replace physicians, but physicians who embrace AI will replace physicians who do not embrace AI (1). The net effect is not “no effect on the workforce”. It will be necessary for pathologists to learn new AI skills they do not have now. It is not the first time pathologists needed to learn new skills to keep up with technology. For most the process of learning these new AI skills will be interesting and enjoyable. This predicted outcome is not a disaster. The patient will benefit immensely from the new precision that
will result. And life goes on.


The Human Cell Atlas – Pathology by the Numbers (2)

Two recent AJT articles refer to the "step change" which the human cell atlas project (HCAP) will bring about in transplantation (3,4). Many readers will want to know how big that step will be. On the one hand HCAP papers are appearing in the very finest journals, and there are videos and articles about it widely available on social media, with predictions of a 10-to-100,000-fold increase in known cell types and claims such as "Our bodies are made up of least 37 trillion cells, and scientists are teaming up around the world to map every single one of them." (5). On the other hand, some large traditional meetings like the US/Canadian Academy of Pathology meeting and the American Transplant Congress have been entirely silent on the issue of the HCAP.

Taking a rigorous approach to the numbers, there is evidence for a least a doubling of known cell types by applying HCAP technologies. Villani et al. (6) have demonstrated a doubling of cell types within dendritic cells, monocytes, and progenitors (from 6 to 12) when HCAP technologies are applied. Extrapolating from that, a conservative estimate of the impact of HCAP is that it will double the number of known cell types in every organ transplanted. So, for instance the number of cell types in the kidney would go from 26 to approximately 52. Even at that level it is something huge that everyone should know about, and be making plans for. It is particularly relevant to the new Banff Classification of Tissue Engineering Pathology, as pointed out in a recent personal viewpoint paper (4).

As for the question of how many cells the project plans to characterize, the HCAP White Paper (7) is quite clear that the intention is to profile 30 million to 100 million cells from healthy controls of both genders in the first draft of the project, and then incorporate the lessons learned from that into creation of a comprehensive atlas of at least 10 billion cells, covering all tissues, organs, and systems. The guiding principle determining how many cells will be analyzed is‘’Given a tissue with N discrete cell subsets, the rarest of which is present at proportion P, how many cells k need to be sampled such that at least n cells are recovered in each subset with confidence level C?”At every step, the HCAP will bring about many important new insights, ultimately
changing and making more precise every aspect of transplantation.

The new Banff Classification of Tissue Engineering Pathology was first suggested in 2011 and concrete plans to make it happen have been in place since 2017, with the aim to have it completely finalized by 2025 (4). HCAP is an integral part of the new Banff Classification of Tissue Engineering Pathology. HCAP should become part of the mind set of every transplant physician and every transplant pathologist, and will need to be central in the joint practical planning of those two communities for the future.


Tissue Engineering Pathology and Regenerative Medicine

It has been gratifying to see the very substantial progress in kidney and liver organoids in the past year (8-11). FDA approval of liver organoids for toxicity testing is expected in 2019. The personal viewpoint paper on tissue engineering pathology and the human cell atlas (4) seems more relevant with each passing month.

– Kim Solez, M.D.


1. Bertalan Mesko, The Medical Futurist , A Guide to Artificial Intelligence in Heathcare, Lean Pub, last updated 2017-11-22,

2. Moghe I, Loupy A, Solez K, The Human Cell Atlas Project by the numbers: Relationship to the Banff Classification Am J Transplant. 2018;18:1830. DOI: 0.1111/ajt.14757

3. Pullen LC. The AJT Report: Human Cell Atlas Poised to Transform Our Understanding of Organs Am J Transplant. 2018; 18 (1):1–2 Last accessed February 8, 2018.

4. Solez, K, Fung KC,Saliba KA,et al. Personal Viewpoint: The Bridge Between Transplantation and Regenerative Medicine. Beginning a New Banff Classification of Tissue Engineering Pathology. Am J Transplant. 2018;18(2):321–327. doi:10.1111/ajt.14610.

5. Weule G. ABC Science – Human Cell Atlas: The plan to map every cell in your body

6. Villani AC,Satija R, Reynolds G, Sarkizova S,…,Regev A,Hacohen NSingle-cell RNA-seq reveals new types of human blood dendritic cells, monocytes, and progenitors.Science. 2017 Apr 21;356(6335). pii: eaah4573. doi:10.1126/science.aah4573. Last accessed February10, 2018.

7. The HCA Consortium, The Human Cell Atlas White Paper, October 18, 2017

8. Brie Przepiorski A, Sander V, Tran T, Hollywood JA, Sorrenson B et al. , A Simple Bioreactor-Based Method to Generate Kidney Organoids from Pluripotent Stem Cells Stem Cell Reports (2018)

9. Nie YZ,Zheng YW,Ogawa M,Miyagi E,Taniguchi H Human liver organoids generated with single donor-derived multiple cells rescue mice from acute liver failure. Stem Cell Res Ther. 2018 Jan 10;9(1):5. doi:10.1186/s13287-017-0749-1.

10. Vyas D,Baptista PM, … Atala A , Soker S . Self-assembled liver organoids recapitulate hepatobiliary organogenesis in vitro. Hepatology. 2017 Aug 23. doi: 10.1002/hep.29483. [Epub ahead of print]

11. Skardal A,Murphy SV, …, Soker S ,Bishop CE, Atala A . Multi-tissue interactions in an integrated three-tissue organ-on-a-chip platform. Sci Rep. 2017 Aug 18;7(1):8837. doi: 10.1038/s41598-017-08879-x.