Your Cells for Sale

Memorial Sloan-Kettering Cancer Center: On the Cutting-Edge of Cancer Care

If you have been a patient at Memorial Sloan-Kettering Cancer Center (MSK), you may be an unknowing participant in a joint venture between MSK and Paige.AI (an acronym for “Pathology Artificial Intelligence Guidance Engine”). [1] Though you will never benefit financially from your contribution to this endeavor, one backed by $25 million in Series A funding primarily from Breyer Capital, you just might benefit in immeasurable ways through the advances in cancer pathology supposedly made possible by the newly-established partnership. [2]

As the first or second-ranked cancer treatment facility in the United States (switching ranks with the University of Texas MD Anderson Cancer Center based on the year), MSK is looked to as the provider of the most cutting-edge and innovative cancer treatments in the country. [3] In recent years, a wave of transformation in cancer prevention, diagnosis and treatment has been started by artificial intelligence companies like Wision AI, which partnered with institutions including Harvard Medical School to develop a “novel deep-learning algorithm [that] can automatically detect polyps during colonscop[ies]” in the colorectal cancer space, [4] and academic institutions like MIT’s Computer Science and Artificial Intelligence Laboratory, which worked with Massachusetts General Hospital’s Department of Radiology to “apply machine learning to the task of distinguishing high-risk lesions that need surgery from those that don’t” for breast cancer patients. [5]

Because of this momentum from other players in the cancer space, in order for MSK to live up to its branding as a provider of “innovative research” and “state-of-the-art science flourishing side by side with clinical studies and treatment,” MSK has no choice but to dive into the machine learning space itself to best manage its cancer care process improvement efforts. [6]

To begin incorporating machine learning into its “strategies to prevent, control, and ultimately cure cancer,” [7] in February 2018, MSK contracted with Paige.AI to provide “an exclusive deal to use [MSK’s] vast archive of 25 million patient tissue slides, along with decades of work by [MSK’s] world-renowned pathologists” in exchange for a 9% equity stake in Paige.AI. [8] Co-founded by MSK’s director of Computational Pathology and chairman of the Department of Pathology, [9] Paige.AI’s mission is “[t]o revolutionize the clinical diagnosis and treatment of cancer through the use of artificial intelligence.” [10]

In the short term, in addition to entering into this partnership, making Paige.AI’s launch possible, [11] MSK will work to provide access to the 5 million digitized pathology slides as well as the 20 million slides which Paige.AI will digitize [12] (while retaining all original slides at MSK). [13] This joint venture also comprises MSK’s plan to integrate machine learning in the medium term, as “[Paige.AI] is years away from selling a finished product.” [14] Once the product is ready, MSK will have a tool capable of suggesting the organ origin, stage and type (primary or secondary) of a patient’s cancer, based on comparing their sample to the 25 million slides provided by MSK. [15] Not only will physicians be better equipped to recommend the most appropriately targeted treatments, MSK will also receive additional funding for research from “[a]ny revenue generated based on MSK’s ownership in” Paige.AI. [16]

As the timeline for Paige.AI’s tool is uncertain, [17] in the short term MSK should begin to test the clinically-validated products that have been developed by other companies, such as SpIntellx’s TumorMapr, which “analyzes the spatial interactions within whole slide images using multiplexed fluorescence . . . to predict patient outcomes to better provide personalized therapeutic strategies,” in order to provide patients the benefits of currently available artificial intelligence tools, even before the MSK-backed Paige.AI system is ready for use. [18]

In the medium term, MSK needs to implement a strategy to incentivize physicians and researchers to develop machine learning capabilities inhouse, rather than starting their own companies outside of MSK. Retaining the product’s life-cycle within MSK is necessary to prevent recurrence of the concerns of stakeholders regarding the MSK-Paige.AI joint venture: patients’ concerns about a third-party profiting from their biological materials without their consent, physicians’ concerns “that the founders received equity stakes in a company that relies on the pathologists’ expertise and work amassed over 60 years” and nonprofit and corporate governance experts’ worries about potential lack of compliance with federal and state law given the absence of a competitive bidding process for exclusive use of MSK’s slides and conflicts of interest created by Paige.AI’s co-founders having influence over MSK’s data resources. [19]

The improvements to cancer care through machine learning could be extraordinary, and nothing could signal this more strongly than a partnership between MSK, a preeminent leader in the fight against cancer, and an artificial intelligence company. However, this venture raises important, and ever more complex, questions about patients’ data. Should organizations be required to solicit patients’ consent for including their information, even anonymized, in the data used to develop machine learning capabilities? Should patients be compensated for their contributions to the tools created? (800)


[1] Paige.AI, “Home,”, accessed November 2018.

[2] Business Wire: A Berkshire Hathaway Company, “Paige.AI Created to Transform Cancer Diagnosis and Treatment by Applying Artificial Intelligence to Pathology,” February 5, 2018,, accessed November 2018.

[3] U.S. News & World Report, “Best Hospitals for Cancer,”, accessed November 2018.

[4] Mike Miliard, “AI Algorithms Show Promise for Colonoscopy Screenings,” Healthcare IT News, November 5, 2018,, accessed November 2018.

[5] Adam Conner-Simons, “Using Artificial Intelligence to Improve Early Breast Cancer Detection,” MIT News, October 16, 2017,, accessed November 2018.

[6] Memorial Sloan Kettering Cancer Center, “About Us,”, accessed November 2018.

[7] Ibid.

[8] Charles Ornstein and Katie Thomas, “Sloan Kettering’s Cozy Deal with Start-Up Ignites a New Uproar,” New York Times, September 20, 2018,, accessed November 2018.

[9] Ibid.

[10] Paige.AI, “Home.”

[11] Ornstein and Thomas, “Sloan Kettering’s Cozy Deal with Start-Up Ignites a New Uproar.”

[12] Ingrid Lunden, “Paige.AI Nabs $25M, Inks IP Deal with Sloan Kettering to Bring Machine Learning to Cancer Pathology,” Tech Crunch, February 2018,, accessed November 2018.

[13] “Memorial Sloan Kettering and Paige.AI,” press release, September 23, 2018, on Memorial Sloan Kettering Cancer Center website,, accessed November 2018.

[14] Ornstein and Thomas, “Sloan Kettering’s Cozy Deal with Start-Up Ignites a New Uproar.”

[15] Lunden, “Paige.AI Nabs $25M, Inks IP Deal with Sloan Kettering to Bring Machine Learning to Cancer Pathology.”

[16] “Memorial Sloan Kettering and Paige.AI,” press release.

[17] Ornstein and Thomas, “Sloan Kettering’s Cozy Deal with Start-Up Ignites a New Uproar.”

[18] “SpIntellx, which Leverages AI for Digital Pathology Markets, Closes Seed Financing and Launches Company,” press release, September 10, 2018, on Tissue Pathology website,, accessed November 2018.

[19] Ornstein and Thomas, “Sloan Kettering’s Cozy Deal with Start-Up Ignites a New Uproar.”

[Image] Paige.AI, “About Us,”, accessed November 2018.


Can ML replace Human Resources?


Open Innovation at General Assembly

Student comments on Your Cells for Sale

  1. I agree that implications of machine learning in pathology and cancer care can be significant! Where I see this having a big impact is in the second opinion space, where today, many/most cancer patients when faced with a concerning diagnosis are suggested to receive a second opinion. The process of sending tumor samples and slides physically to different sites, let alone ensure there is enough tissue to run tests/review is painful and incredibly inefficient. Machine learning can not only help validate or quicken the diagnosis process, but the digital pathology platform for this can be used for so many applications beyond AI. To be able to share tumor slides digitally across institutions and even countries can be a saving activity when cancer treatment decisions and ultimately someone’s life is on the line.

  2. This is a fascinating piece about the promise of machine learning in advancing cancer treatment. In order to assess the kind of magnitude of information that exists in the cancer pathology space, I believe this is one of the most innovative, smart, and forward-looking initiatives one could do, as provocative or controversial as it might be. You bring up excellent points about the potential for patients to feel disenfranchised by this, however I would say that one of the reasons that patients often agree to participate in clinical trials, for example, is that their participation will benefit them (potentially) but also that it may benefit people in the future (their children, their children’s children, etc). I think this is a risky endeavor for all of the reasons you point out, but the promise is huge, and if the messaging to patients is correct, it has the potential to bring enormous benefit to cancer patients for years to come. Thank you for this insightful piece!!

    By the way, you might also take note of this scandal at MSK that just happened recently, where one of their most senior leaders failed to disclose his kick-backs from pharma, further pointing toward the idea that MSK has an uphill battle to fight here:

  3. Thank you for the insightful and well-written article! I really learned a lot.

    After reflection, I believe this is a gross infraction on human privacy and miscalculation of risk.

    What is not discussed in this article (and rarely discussed in big data-driven consumer applications to healthcare, defense, etc.) is the vulnerability of data. Given the sensitive nature of key information like a person’s health information, ventures like Paige.AI highlights how risk is not properly measured given the ambition of the firm to generate out-sized returns. The risk of this data being breached is wide-ranging, impacting all stakeholders, including (and most importantly) the clients. All for what – to develop drugs that will be priced beyond the means of the person who needs it? Using third-party vendors and not compensating or allowing customers to provide consent for their data usage are just two examples of ways this venture is not properly paying for the risks it takes.

    While I understand how data needs to be collected unilaterally to avoid selection bias, there has to be a form of compensation and disclaimer to all patients who are served in this program. To do anything less is to expose people to risks unimaginable; if hacked, people can change their website passwords, credit cards and even banks accounts, but one cannot change their genetic code and/or other identifying health factors unique to a single person.

Leave a comment