Artificial intelligence in cancer therapy

One in two people in Germany will develop cancer at some point in their lives; there are around half a million new cases every year. Fifty years ago, a cancer diagnosis was still tantamount to a death sentence for the majority of those examined, but thanks to advanced treatment options, two-thirds of those suffering from the disease are now cured. Recently, artificial intelligence has also been used to diagnose and treat cancer. Will AI (once again) revolutionise cancer therapy? There is still a long way to go, but AI is already an interesting and promising new addition to the medical toolbox.

Numerous research projects are currently trying to make AI useful for cancer therapy. This text presents a few approaches. The overview is not representative, but it does give an impression of the range of possible applications.

AI in the diagnosis of cancer

The ability of artificial intelligence to recognise structures in large amounts of data is increasingly being used in medicine for cancer diagnosis. In fact, doctors today have more data at their disposal than in the past. For example, there are more and more imaging examinations: computed tomography and other X-ray procedures, magnetic resonance imaging, ultrasound examinations, microscopic examinations of tissue, molecular genetic examinations or skin screening with visual documentation.

Artificial intelligence can provide valuable services in the analysis of these images. AI models are trained with case studies and can then recognise early forms of cancer in current patient images. This works with different types of images. In Dresden, for example, an AI developed at the Technical University detects genetic changes in samples of intestinal tissue. Another procedure carried out at the university hospital is the analysis of images from mammography, in which the AI identifies suspicious lumps that could be early stages of cancer. At the Charité hospital in Berlin, AI integrates information from various imaging examinations into an overall picture.

Complex data is also generated during genetic analysis. Here, too, AI can help: in a study conducted by Charité and Ludwig Maximilian University of Munich, for example, artificial intelligence was used to detect modifications in the genetic material of tumours – this made it possible to identify different types of nasal and paranasal sinus tumours.

Such methods can be used not only to identify early forms, but also pre-forms of cancer or mere genetic risks (predictive analytics). This means that preventive measures can be taken before cancer has actually occurred.

AI in the development of cancer therapies

In the development of cancer therapies, complicated considerations have to be taken into account, which incorporate the previously collected examination data. Artificial intelligence can help to shorten this process.

The Cancer Scout project in Göttingen, for example, uses AI to digitally analyse tumour tissue samples (digital biopsy) and assess whether a tumour is likely to respond to personalised therapy. Only positive findings will be followed up with a detailed laboratory examination. The digital biopsy replaces large-scale molecular testing, which would not be affordable due to the enormous amount of time and resources required.

Couldn't AI also independently develop therapies for a patient? After all, not only patient data is available in digital form, but also scientific studies on the current state of research. In fact, Charité has conducted a study in which large language models like ChatGPT were used to develop therapy recommendations for specific cases – at the same time, experienced doctors were also consulted on the same case. The AI results were sometimes amazingly good; but in most cases, they were inferior to the medical recommendations. 

Supporting therapies

For the foreseeable future, AI will not make doctors redundant. In the development of therapies, it remains a tool that supports human experts. This also applies to the implementation of therapy decisions. Here, too, artificial intelligence is used in various sub-areas. Two examples are given below:

In radiation treatment, AI helps to localise the irradiation site so that as little surrounding tissue as possible is damaged. A new method developed in Berlin that links a radiation device with a computer tomograph is used for this purpose. AI identifies the current position of organs in the body and the structure of the tumour in real time; it uses these findings to make suggestions regarding the approach of the irradiation.

In surgical removal of tumours, AI can be used in robot-assisted surgical systems to perform more precise operations.

AI to support patients

But artificial intelligence can do more than just make doctors' work easier – it also helps patients to adjust to their treatments.

For example, AI-supported patient information systems run through therapy plans and present opportunities, risks or side effects over time. This gives patients a better basis of information for their decisions about therapies. AI-supported apps and digital measuring devices in patients' possession can monitor health in real time and transmit changes to treating physicians.

Problems with the use of AI

It has already been mentioned: AI models for analysing patient data always have performance limitations, the nature of which depends on the one hand on the model used and on the other on the quality of the learning data. Better models and better data should gradually reduce these deficits. However, there are also more fundamental problems.

For example, AI models are not transparent. People are usually unable to understand how an AI has arrived at an assessment. If a statement proves to be false, it is almost impossible to determine why. It is difficult for humans to learn from AI mistakes.

AI results are also not always reproducible. Anyone who has worked with language models like ChatGPT is familiar with the phenomenon: the same query is sometimes answered differently at different times or in different content contexts. What does this say about the quality of the AI assessment?

Many doctors currently have only limited trust in the assessments of artificial intelligence, and that is probably a good thing. However, this does little to detract from the importance of AI. Just as a tool cannot build a house without the help of a qualified craftsman, the instrument of AI also requires an experienced team of doctors to help heal a patient.

Ultimately, research is still in its infancy. As astounding as it may seem, the full potential of artificial intelligence for combating the widespread disease of cancer is still far from being foreseeable.

Text: Thorsten Kleinschmidt