Artificial Intelligence in Oncology: Fantasy or Reality?

Artificial intelligence promises to use the power of data to solve some of the biggest problems of our time. But can it help us treat a disease as complex as cancer?

Artificial intelligence and machine learning are (not that) new technologies that have been recently boosted thanks to hardware improvements. Through algorithms, they can learn, predict and advise based on vast amounts of data.

Its potential to disrupt all sorts of markets has led to some big investments. In April, the European Commission announced a €20Bn AI strategy for Europe. France also launched its own €1.5Bn program, which was followed by the opening of new R&D facilities by companies like Fujitsu, Facebook or Google DeepMind.

One of the areas where AI is expected to have a major impact is healthcare, where it can be used to interpret the data from the massive databases gathered over the years by companies, healthcare providers and payers. In particular, the treatment of cancer could greatly benefit from the arrival of AI technology.

Why is artificial intelligence relevant in oncology?

Oncologists have been trying for decades to define small subsets of cancer patients that can benefit from a specific treatment. However, the success of targeted therapies has so far been limited. At the moment, medical doctors are overcrowded with data from imaging, genomics, co-morbidities and previous treatments.

This is where AI comes into play. The technology has the potential to crunch the data to predict the prognosis of the patient and advise doctors with different options available, including personalized medicine and clinical trials with experimental therapies.

AI as a diagnostic tool

Some companies are already selling ‘AI as a service’ solutions ranging from early stage diagnosis to prognosis. For example, in the context of breast cancer, only 5% of women who are recalled after a first screening are indeed sick. This increases the costs and is a stressful experience for patients. Therapixel, a startup specialized in medical imaging, is using artificial intelligence to deal with this issue by performing automated mammography analysis.

In the specialty of pathology, AI has shown that it can significantly reduce the error rate of diagnosis as compared to a specialist on their own. In this field, Google is developing an augmented reality microscope that uses AI software to assist pathologists in the detection of cancer, which could reduce significantly some time-consuming activities such as manual cell counting. IBM has ambitious goals with its AI product Watson for Genomics, although so far its results don’t seem as good as promised.

In Switzerland, Sophia Genetics is using artificial intelligence to pinpoint gene mutations behind cancer to assist doctors in the prescription of the best treatment. Their solution costs on average $50-$200 per genetic evaluation and according to the company it is currently used by more than 420 hospitals in over 60 countries.

Another deep tech innovation for early detection of cancer from Freenome has attracted a $77M round from well-known VCs, including Andreessen Horowitz and Google Verily. Freenome recently announced a strategic collaboration with the Institut Curie to evaluate its AI genomics platform as a tool to predict patients’ responses to immuno-oncology therapies by observing changes in biomarkers circulating in the bloodstream.

AI for precision medicine

Using AI to stratify patients has big potential, but a major bottleneck is that we are still lacking a range of personalized medicine drugs wide enough to treat all these patients. According to Sam Natapoff, analyst at Bloomberg, drug development is “made for AI applications.” This opportunity has attracted large AI developers, big pharma and a huge number of startups. It is estimated that approximately one hundred startups are using AI in the field of drug discovery.

In late 2016, Pfizer announced a collaboration with IBM Watson for Drug Discovery in order to “analyze massive volumes of disparate data sources, including licensed and publicly available data as well as Pfizer’s proprietary data.”

Sanofi and GSK have announced, respectively, $300M and $42M deals with Exscientia, a spin out of the University of Dundee, Scotland, to identify synergistic combinations of cancer targets, to then develop drugs against those targets.

Roche, out of many other deals including the acquisition of Flatiron Health for $1.9Bn and a partnership with GNS Healthcare, is supporting an open research initiative called EPIDEMIUMto bring together multiple players and apply AI to the research of new cancer therapies.

However, this field is still at a very early stage. So far, only the British company BenevolentAI, in partnership with Janssen, has shown concrete results, which have led to a drug candidate now moving to a Phase II trial.

Reducing trial costs

Artificial intelligence has the potential to draw insights from tremendous volumes of real-world data and apply it to the design of clinical trials, which could reduce significantly the cost. Especially given that patient recruitment alone represents about 30% of the total clinical trial time.

Recently, the Horizon 2020 program granted €16M to a huge European consortium — including big names like Institut Curie, Charité, Bayer, Philips and IBM — aiming to use AI technology to improve clinical outcomes in oncology at lower cost.

However, precedents have not been that promising. In 2013, the M.D. Anderson Cancer Center launched a program to test whether IBM Watson could speed up the process of matching patients with clinical trials. In the end, the $62M program didn’t prove to be efficient and cost-effective.

Challenges to overcome

Data scientists have to deal with unstructured electronic health records, and data coming from multiple sources that has been collected and structured for different purposes. Most routine databases do not have sufficient quality to be used by AI algorithms to achieve the quality standard required for clinical trials.

From a regulatory perspective, the authorities have been proactive to address issues in the approval process. FDA Commissioner Scott Gottlieb said recently during a conference in Washington“AI holds enormous promise for the future of medicine,” and “We’re actively developing a new regulatory framework to promote innovation in this space, and support the use of AI-based technologies.”

The very first cloud-based deep learning algorithm has been approved recently by the FDA under the category of medical devices, meaning it can be used in clinical routine. In the EU, there is a legislative proposal for new medical device regulations, not yet adopted, that addresses software for medical devices with a medical purpose of “prediction and prognosis.” An additional challenge in Europe is that the recent General Data Protection Regulation (GDPR) enforcement is also impacting the development of AI algorithms.

There is undoubtedly fantasy around AI. Entrepreneurs are tempted to surf on the hype and the limited understanding of the community about what AI is and what it can do. The AI business value chain should be discussed to clarify the involvement of different stakeholders in all steps, from raw algorithms to results. That way, we will be able to finally switch from an overhyped technology with only a few proof-of-concept examples to a breakthrough in healthcare.




Συνδεθείτε για να δημοσιεύσετε το σχόλιο σας:


Σχολιάζετε χρησιμοποιώντας τον λογαριασμό Αποσύνδεση /  Αλλαγή )

Φωτογραφία Google

Σχολιάζετε χρησιμοποιώντας τον λογαριασμό Google. Αποσύνδεση /  Αλλαγή )

Φωτογραφία Twitter

Σχολιάζετε χρησιμοποιώντας τον λογαριασμό Twitter. Αποσύνδεση /  Αλλαγή )

Φωτογραφία Facebook

Σχολιάζετε χρησιμοποιώντας τον λογαριασμό Facebook. Αποσύνδεση /  Αλλαγή )

Σύνδεση με %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.