AI – a game changer in cancer

This question was asked at a recent conference titled Intelligent Health, organised by Oslo Met, Oslo Cancer Cluster and Akershus University Hospital. The event brought together over 200 live participants and 100 on streaming, including students, academics, and practitioners from across a number of institutions and companies, with the purpose of discussing, understanding, and sharing.

The talks covered a wide range of topics, from using AI in the development of personalised health, to prevention and treatment of different diseases. Several presentations focused on AI solutions in the oncology sphere, including prevention, drug development and treatment optimisation.

Watch the recording of the conference here: The Intelligent Health conference 2024 – FilMet (oslomet.no)

Creating digital twins

Arnoldo Frigessi, professor at the University of Oslo, presented Digital twin for personalized treatment: example from breast cancer. Frigessi outlined an “in-silico approach” to personalised treatment. By using all available data of the patient, he described how to make a digital twin and simulating different treatment options.

“The fundamental idea is to produce copies of the unique patients on a computer. When we have a lot of copies of the patient we can treat each copy with a drug or a different dose and compare them in-silico on a computer to see which one of the drugs works better. This is the in silico approach to personalised treatment,” Frigessi stated.

The simulation is still far from being used in a clinical setting, due to time constraints, but gives a view of what personalised cancer treatment may look like in the future.

Improving screening strategies

Oluf Dimitri Røe, professor at the Norwegian University of Science and Technology, covered the clinician’s perspective: The role of AI in predicting lung cancer. Røe introduced how a new machine-learning model can improve the selection of participants to lung cancer screening.

“Yearly screening with CT scan could reduce the mortality of lung cancer with 20%, however, in the NLST study they found that only 26% of everyone that got lung cancer had been included in the study. This was due to the smoking criteria they used. So we need new models for selecting people for lung cancer screening.

“One of the machine learning models is the Hunt Lung Cancer Risk Model, which was published in 2018. We used backwards feature selection on 36 variables that we picked out from the Hunt data bank. The main thing is that we could with a simple calculator get a very high AUC value, higher than the PLCO in our population,” commented Røe.

Predicting treatment outcome

Tero Aittokallio, group leader of Computational Systems Medicine in Cancer, Dept. of Cancer Genetics, Oslo University Hospital gave the researcher’s perspective: AI for treatment optimization in pancreatic cancer and hematological malignancies. Aittokallio talked about building multimodal AI to predict treatment outcome in pancreatic cancer.

“This is based on data that we collected in a Horizon 2020 project for the last four years. Most of the time has been spent collecting data from different hospitals in Europe, such as liquid biopsies and surgery. We are using genomics, pathology, MRI imaging, radiology, and then the idea is that we are combining this data using AI models and using this information to select the right treatment for the individual patients.”

Identifying new drug targets

Amine Namouchi, Principal Scientist in Nykode Therapeutics, represented the industry’s perspective with The role of AI in creating personalized cancer vaccines. Namouchi described how Nykode has developed AI models to detect novel drug targets for development.

“For the case of cancer vaccine development, our starting point is of course the patient. From the patient, we have a blood sample and a tumour biopsy sample. We perform the DNA and the RNA sequencing, and HLA typing, to know exactly what MHC class I are expressed at the surface of this particular patient. Then we apply our method and our algorithm to select and rank those neoantigens, and the platform is called NeoSELECT,” said Namouchi.

Namouchi then described how the company identifies the neoantigens that bind to the MHC molecule, by using an artificial neural network.