How AI Will Push The Frontiers Of Modern Medicine

Sophia Haegerich

Today, the United Nations estimates that our globe is home to over 7 billion people. That’s seven times more people than only 200 years ago. And on top of this rapid population growth, according to OurWorldInData, we are living longer.

However, what seems to be great news at first poses some huge challenges for current and upcoming generations. For instance, Eurostat found out that while population and average span of life have increased, the years spent in good health have not. This means a lot more people will rely on several extra years of medical support. Especially, degenerative diseases like dementia will become growing issues. For instance, the World Health Organization (WHO) estimates that the number of people living with dementia will increase from currently about 47 million to 75 million by 2030 and triple by 2050.

The projection of such stark reality raises a number of questions about the future of healthcare. How will we deal with the additional patients and overcome the obstacles of providing care as costs and needs increase? And how will the role of doctors change with this development?

Machine learning is pushing the frontiers of modern medicine

Clearly, diverse actors beyond the medical domain must combine their expertise to find solutions to such questions. The emergence of new technologies is an important element to consider during their discussion. Here, machine learning, a branch within the area of artificial intelligence (AI) helping computers to learn from data and discover patterns without being explicitly programmed, is an essential force in steering change. But what has AI achieved in medicine so far and what will be possible within the next decades?

Status quo of AI in the medical domain

While there has been an immense growth of medical knowledge and progress in the area of artificial intelligence, so far only a minor fraction of AI research has focused on healthcare. Only recently have data pools necessary to apply machine learning become publicly available. The reasons for this late availability include patient privacy regulations, error-prone or imprecise data, as well as problems integrating scattered data from several disconnected sources.

With the growing Big Data trend and recent developments in precision medicine, an emerging concept to prevent or treat diseases considering a patient’s variability in genes, surroundings, and lifestyle, comes an increasing awareness of the value of diverse medical data.

In recent times, the field of medical image recognition has experienced major progress thanks to new machine learning techniques. While the technology of deep learning has its roots already in the 1960s, just recently has its full potential became noticeable, due to, for instance, increased data access, algorithmic improvements, and advancements in parallel data processing.

Although humans are strong in pattern recognition for low dimensions and still exceed machine performance by far, manual analysis has become much more difficult for doctors. This has led to the creation of several new positions in the hospital setting, which strongly differ from the traditional roles one normally thinks of.

A good example is the case of Prof. Dr. Christian Wachinger, a computer scientist leading the Laboratory for Artificial Intelligence in Medical Imaging at the child and adolescent psychiatry of Ludwig-Maximilian-University (LMU) Munich. He stresses that “the need for data analysis creates a variance in the hospital setting and opens up new possibilities for computer scientists and data analysts. In short, there is a growing need for people with a unique ability to find meaning behind rising medical data sets.”

The interim professor at the university hospital Munich, who works together with multi-disciplinary teams coming from the medical, technical, and psychological domain, uses machine intelligence to solve clinical problems, such as psychiatric disease or brain anomaly. One recent example of their work is the Brain Age Project carried out in collaboration with SAP. For the project, the team of LMU Munich was joined by the SAP machine learning research team, which works together with diverse academic partners to discover new machine learning methods and trends to empower SAP’s enterprise solutions.

Brain age project

Located on hospital premises, the LMU team is at the source to support doctors and patients with their work on advanced treatment methods. This led them to a new framework based on age estimation in the neuroimaging field, which was executed together with machine learning researchers from SAP.

Wachinger explains that the idea of the research project came up when the team discovered that the manual interpretation of magnetic resonance imaging (MRI) scans of the brain was becoming more complex and time intensive. This increased difficulty for doctors is due to many factors, such as larger data sets or growing resolution, and created the need for advanced analysis methods. At the same time, the analysis of MRI scans is of great importance since anatomical changes in the brain can be used to predict a decline in a patient’s cognitive abilities.

Therefore, the Brain Age Project team has used around 1,000 anonymized publicly available MRI scans of healthy subjects to train a machine learning model that spots the signs of aging in the brain. These MRI scans built the basis to detect anomaly in patients when estimating the actual brain age and comparing it to the patient’s chronological age. Hypothetically, a healthy patient should not have a significant difference between both ages. However, for patients with progressed dementia prior to showing symptoms, the model should determine a larger difference between biological and chronological age, allowing doctors to depict signals of disease in its early stage, explains Wachinger. He further stresses that an early detection of neurodegenerative diseases is important because it allows the application of therapies that slow down the progression before symptoms are visible.

Machine learning can help physicians make earlier diagnoses of dementia

The project’s underlying process can be compared to a blood test, where the doctor takes a blood sample and sends it to an external laboratory to get results, which he or she then checks to see whether they lie within a normal range. In the Brain Age project this process is similar: The doctor puts the patient in a scanner to receive an MRI while the machine’s role resembles the role of the laboratory. It uses the MRI scan to run the defined algorithms and provides doctors with a precise analysis of the brain structure.  The doctor then uses this set of results to compare it with the predefined parameters of a healthy subject in order to make a diagnosis. For the patients, this proposed process also comes with advantages. They are still examined and diagnosed by a doctor, and can count on a highly accurate analysis of their brain structure, as well as a shorter waiting time from examination to diagnosis.

Intelligent machines and the future of healthcare

While the Brain Age Project shows the potential of AI in healthcare, the joint efforts of the LMU and machine learning researchers from SAP only mark the beginning of what will be possible in the future. In a future scenario, the developed deep learning algorithm could provide doctors with an automatic analysis of the brain structure while the patient is still in the scanner. This would save the physician plenty of time, which he or she could use for more relevant tasks, such as patient treatment. Until then, ongoing research, testing, and measurable improvements will be necessary to pave the way for intelligent machines in the hospital setting. These machines will not replace doctors, but support them in their everyday work.

Wachinger also sees the role of machine learning in the medical domain as well positioned: “On the one hand, you could try to help the doctors to do their job more efficiently. On the other hand, you can do new things that couldn’t be done before.” These improved and new treatment methods can make a valuable contribution to answer the increasing healthcare challenges of our growing and aging society.

As AI technologies develop, they will transform the way doctors examine their patients, enlarge the possibilities to predict and cure diseases, save healthcare costs and improve medical care in regions where access to healthcare is scarce. Finally, picturing a future of medicine based on data and analytics gives reason for hope but requires continuous research to realize its full potential. The basis of this research is formed by machine learning researchers like Wachinger, who work with diverse medical data to “improve patient care and gain a better understanding of disease.”

Learn more about the Brain Age Project:

For more information about SAP Machine Learning Research, visit medium.com/sap-machine-learning-research.

This article originally appeared on SAP News Center.


Sophia Haegerich

About Sophia Haegerich

Sophia Haegerich drives the communications of the machine learning research team at SAP. The team works on solving tough machine learning problems for multiple applications, scenarios and platforms. The potential impact of these general machine learning models on a wide range of use cases across finance, procurement, logistics, or travel management is important to enable the intelligent enterprise. Sophia dedicates her time to spread the word about her team’s research collaborations with academia, as well as the new machine learning trends and methods discovered along the way.