
.
All aboard the AI train
Although this wave of popularity is certainly pleasant and exciting for those of us working in the field, it carries at the same time an element of danger. Information theory is indeed a valuable tool and will continue to grow […], but it is certainly no panacea for the communication engineer or anyone else. […] It will be all too easy for out somewhat artificial prosperity to collapse overnight when it is realized that the use of a few exciting words like information, entropy, redundancy, do not solve all our problems.
- Claude E. Shannon, "The Bandwagon", 1956
Since its earliest days, Artificial Intelligence (AI) has often been over-promising but under-delivering. More than fifty years later, Shannon’s concerns are still very timely: just switch ‘information theory’ with ‘machine learning’ and ‘ information, entropy, redundancy‘ with ‘deep learning, neural networks, AI‘ and you’ll suddenly be projected from 1950 to 2020.
Nowadays, everybody wants to _make AI._ Big-tech companies come out with more complex and resource-demanding neural networks every year. Money flows, the hype rises, and even quite conservative branches like Healthcare start going with the flow without worrying enough about its implications.
Concerning AI, Deep Learning (DL) has been the real M.V.P. of this decade, recently establishing as a real value-adding technology in different fields of science and pushing itself into the spotlight as the closest thing we have to human reasoning.
Except we’re not there yet.
Hereby we will discuss the reasons why, even if the hype-powered Deep Learning bandwagon can’t and must not be stopped, Healthcare providers should dwell more on who may really benefit from this new approach.

AI and Healthcare: a better love story than Twilight
The relationship between Medicine and Artificial Intelligence is as old as the AI field itself. In this sense, we can say that AI found in Healthcare the first, __ big _crus_h of its youth.
Early days and first breakup
The earliest medical expert systems (i.e. computerized problem-solvers made of hardcoded if-else statements) can be in fact traced back to the beginning of 1960. As every teenage love story, the first moments had been memorable and the enthusiasm was at its peak, but things started to get a bit complicated quite soon: due to hardware limitations, lack of foundings and an abundant dose of pessimism from both the academics and the press world, Healthcare – and many other science branches – progressively lost interest in AI and its application. The first breakup was in the air.
Recovering
The so-called AI winter lasted until 1990. In the meantime, researchers focused on how to use expert knowledge to achieve clinical advice for patient care based deductive learning. These were also the years when Judea Pearl and Richard Eugene Neapolitan came up with the idea of Bayesian Networks, one of the most charming statistical models that pioneered Clinical Reasoning and still has a lot to say even 45 years after its creation. Bayesian classifiers and Bayesian theory in general become soon a well-established practice to pair with the good old Regression Models, leading to huge improvements in many fields of medical research for at least two decades. Then the Internet happened.
Back together
Starting from the mid-90s, the amount of data collected began to increase exponentially, along with the impressive improvement in computers’ hardware and computational power. A decade later, massive amounts of data started becoming available also for the medical domain with the spread of EHRs (Electronic Health Records). Soon, healthcare researchers began shifting from a knowledge-driven approach to a data-driven approach: the romance was starting again.
![ML Publications in PubMed in the last two decades [Valliani et al. 2019]](https://towardsdatascience.com/wp-content/uploads/2020/02/1Ng7SMwWUjQy_WHb7MJO2zA.png)
Until data do us part
Fast-forwarding to the 2010s and the Big Data revolution, a "new" fancy AI technology shook the scientific community with its impressive achievements. Deep Learning algorithms started breaking records one after another, gradually settling various flavors of Artificial Neural Networks (ANNs) as the state-of-the-art technology in fields like Natural Language Processing, Computer Vision and Speech Recognition, leading to important advancements also in Medical Imaging.
![The exponential growth of Data and Deep Learning in Healthcare [Fang et al. 2016 (R), Beers et al. 2018 (L)]](https://towardsdatascience.com/wp-content/uploads/2020/02/1ZZv-jst00AWdN2nD3682Yg.png)
The Two Sides of Deep Learning
Why everyone loves Deep Learning?
Contrary to traditional Machine Learning (ML) algorithms, Deep Learning is fueled by massive amounts of data and requires high-end machines with powerful GPUs to run within a reasonable timeframe. Both of these requirements are expensive, so why do companies and research labs think the juice worth the squeeze?
In traditional Machine learning techniques, most of the applied features need to be identified by a domain expert in order to reduce the complexity of the data and make patterns more visible to learning algorithms to work. The biggest advantage of Deep Learning algorithms is that they try to learn high-level features from data by themselves. This theoretically eliminates the need for domain expertise and hardcore feature extraction.

In complex problems where a high level of automation is required but there is lack of domain understanding for feature engineering (e.g. Image Classification, Natural Language Processing, Robotics) Deep Learning techniques are skyrocketing reaching never-seen-before levels of accuracy.
Deep Learning Applications in Healthcare
By processing large amounts of data from various sources like radiographic images, genomic data and electronic health records, Deep Learning can help physicians analyze information and detect multiple conditions, trying to address many needed healthcare concerns like reducing the rate of misdiagnosis and predicting the outcome of procedures. Here there are some well-known medical areas where Deep Learning is currently showing off:
- Tumor detection: the adoption of Convolutional Neural Networks (CNNs) significantly improves the early detection of cancer, reaching very high accuracies in problems like breast cancer detection on screening mammography [Shen et al. 2019]. In this field, DL algorithms are approaching – or even surpassing – the accuracy of human diagnosticians when identifying important features in diagnostic imaging studies [Haenssle et al. 2018].
- Hospital readmissions, length of stay and inpatient mortality forecasting: DL-powered algorithms have access to tens of thousands of predictors for each patient, including free-text notes, and automatically identifies which data are important for a particular prediction without hand-selection of variables deemed important by an expert [Rajkomar et al. 2018].
- Drug discovery and precision medicine: the discovery of a new drug is always surrounded with excitement from the academic community and the general public, but the drug development cycle is very slow and expensive, and less than 10% make it to market. DL can be used to automatically produce fingerprints and more effective features or for de novo drug design, reducing the cost of the process [Xu et al. 2018].
- Natural Language Processing: the introduction of EHRs in medical centers all around the world unlocked a new source of information to leverage for healthcare providers: free text. Extracting useful actionable information from unstructured data helps in many aspects of healthcare, like summarization, automated reporting, question answering and – of course – decision making. However, clinical text is often fragmented, ungrammatical, telegraphic and make heavy use of acronyms and abbreviations, which can stump even the smartest NLP algorithms.
However – as the title says – even in Deep Learning, not all that glitters is gold. ** Data scientists keep putting a lot of effort in all of the aforementioned applications, but some of them expose limitations that seem too hard to overcome in real use-case scenarios, forcing DL-based methods to stay relegated in the research-only quarantine zone. But why these methods excel in some areas and struggle in others? Why do we strive to achieve significant progress in real use-case scenarios**?
Here we will focus on two of the most important conceptual issues of Deep Learning in Healthcare.

Deep Learning & Healthcare Issues: Interpretability and Causality
Dealing with Healthcare means dealing with people’s life. This implies carefulness, confidence, transparency, caution, sensibility and the ability to explain why and how we end up with a certain diagnosis. In the same way we expect to find these qualities in physicians and surgeons, we should seek for them also in our algorithms. And here is where Deep Learning shows its limits.
Let’s take Natural Language Processing, for instance. Nowadays, the amount of human-generated written/spoken data in healthcare is massive. It’s been estimated that nearly 80% of the healthcare data remains unstructured and untapped after it is created. Clinical notes are probably the most ignored input in healthcare, and this happens not because they’re not informative but just because it is hard to handle this type of data. We already mention how leveraging this kind of information can lead to solid improvements in models’ accuracy, but is performance all that matters?
A very popular way to process text for prediction in healthcare is to use word embeddings, a multi-dimensional, dense, numeric representation of words powered by various types of Neural Networks. Patient’s clinical notes can be then combined in different ways retrieving document embeddings or patient embeddings. Since we’re dealing with numeric vectors instead of text now, we can simply feed them into our favourite classification algorithm.

Let’s suppose now we developed a model for the early detection of cancer that sees an astonishing 30% accuracy boost when we include unstructured data through 300-dimensional patient embeddings. We reasonably guess then that our model is using some very relevant information within the clinical notes to assess the patient’s condition.
But what is this relevant information?
As we said, embeddings are just dense numeric vectors. Converting words into vectors using Deep Learning and merging them together completely shuffles the cards on the table, making impossible to guess the combinations of words/sentences responsible for the patient’s classification. Attention mechanisms and coefficients can only tell us what are the most relevant components for predicting the outcome, but since we lost the connection between words and components of the embeddings we cannot really understand why these components are so important.

Even if the model reaches a better accuracy, we can’t diagnose a desease just because component #217 says so. Using dimensionality reduction is fine, but we must be able to revert the process if we need to understand the origin of the decision taken and evaluate it within a bigger (human) knowledge framework. Which leads us to the causality issue.

Yoshua Bengio, AI pioneer and Turing Award winner for contributions to the development of Deep Learning, recently states that DL has to learn more about cause and effect in order to achieve its full potential. In other words, he says, Deep Learning needs to start asking why things happen.
Deep Learning is fundamentally blind to cause and effect. Unlike real doctors, deep learning algorithms cannot explain why a particular image or clinical note may suggest disease. Understanding causation, especially in healthcare, would make existing AI systems smarter and more efficient: a robot that interiorize causal realtionships can formulate hypothesis, imagine scenarios and learn how to address them before they actually happen.
This mindset is peculiar to the human nature, and cognitive science experiments have shown that understanding causality is fundamental to human development and intelligence, although it’s still unclear how we form this kind of knowledge. Deep Learning algorithms, on the other hand, aren’t very good at generalizing and fails in applying what they learn from one context to another. This is very limiting in medicine where every diagnostic and prognostic task is based on a very complex network of causal connections.

Conclusions
What we got so far
Deep Learning has been a game changer over the last decade, and we’ve seen the many ways it can be implemented in the medical field. Nevertheless, we must point out that its shady and narrow nature is a problem when we deal with very complex and delicate situations involving patients’ lives. Having an extremely accurate pattern recognition tool is useless in clinical decision making if we can’t unwrap it to understand the way it gets its conclusion.
That’s why simple linear regressions still win over complex neural networks in many real use-case medical scenarios. We can’t trust pseudo black-boxes that can’t be fully explained, in the same way we can’t rely upon highly specific models that can grasp correlation but not causation if causation is the keystone of the discipline.
What the future holds
In the epilogue of his latest book, Reeboting AI, Gary Marcus says:
"AI that is powered by deep understanding will be the first AI that can learn the way a child does, easily, powerfully, constantly expanding its knowledge of the world, and often requiring no more than one or two examples of any new concept or situation in order to create a valid model of it."
This is arguably the next big achievement we should point to, but it won’t be easy. Creating an intelligence that exceeds human levels is far more complicated than we have been led to believe. The world is complex and open-endend, and the highly specialized narrow machines we have today can’t be completely trusted yet for those tasks where this complexity is fully displayed, as in personalized medicine.
References and further lectures
Here you can find my main sources, along with some very interesting readings which I suggest you to have a look at if you are interested in digging deeper into the complex relationship between Deep Learning and Healthcare.
- A Meta-Transfer Objective for Learning to Disentangle Causal Mechanisms, Bengio et al. 2019
- An AI Pioneer Wants His Algorithms to Understand the ‘Why’
- Human Compatible – Stuart Russell
- Rebooting AI – Gary Marcus and Ernest Davis
- Causality – Judea Pearl
- The balance: Accuracy vs. Interpretability
Hopefully this blog post has been able to give you an overview on the many implications of using Deep Learning in medicine. Please leave your thoughts in the comments section and share if you find this helpful!
Tommaso Buonocore – Author – Towards Data Science | LinkedIn