Multimodal disease prediction and AI

The Hjaltelin group applies linkable temporal multimodal health data from Danish registries, genetics, and beyond to develop machine learning/AI methods for personalised health care. The group's overarching goal is to bridge machine learning/AI methods with clinical application to develop health care options, such as screening and treatment, more tailored to each individual patient.

The group is highly collaborative within-department, with international research teams, and clinical experts to help translate results into clinical practice. Also, we are used to working in an interdisciplinary environment, spanning bioinformatics, statistics, machine learning/AI, and computer science.

Research focus

The Hjaltelin group develops machine learning and artificial intelligence models for:

1) early detection of complex diseases

2) personalized patient outcomes

3) the translation of these insights into clinical practice.

The overarching goal is to pioneer personalized screening and treatment strategies by leveraging individual patient risk and explainable AI approaches.
The group works with multimodal data and longitudinal patient histories, including diagnoses, medications, lab measurements, and primary care events, but also linked molecular data such as genetics and protein levels.

A core focus is on explainability and visualization for machine learning models to enable better explorative analyses, interpretation, and communication of results - such as early symptoms and temporal risk factors. The group collaborates closely with clinical experts in a multidisciplinary environment to effectively translate findings into real-world clinical practice, ultimately improving patient outcomes.

Main findings

The Hjaltelin group has developed an AI-algorithm, in collaboration with the Brunak group, capable of detecting pancreatic cancer early on using the nationwide Danish National Patient Registry (comprising 8 million) patients, and cross-validation on 3 million American patients:
- Nature medicine: Nature Medicine article
- DR: DR article
- Videnskab.dk: Videnskab.dk artikel

Furthermore, the Hjaltelin group works on multimodality data and advanced machine learning models for broader cancer detection and risk assessment as explained in our review:
https://www.thelancet.com/journals/lanonc/article/PIIS1470-2045(24)00277-8/abstract

The group also has a high emphasis on systems-level understanding of comorbidity and multimorbidity, exploring how these can impact disease progression and patient outcomes:
- Nature Reviews Genetics: https://www.nature.com/articles/nrg.2016.87

Projects in the group

Early detection of pancreatic cancer using multimodal patient trajectories and AI

Pancreatic cancer is one of the hardest to detect early, with a 5-year survival rate of just 12%. This project leverages artificial intelligence and Danish health registries to uncover links between pancreatic cancer and new-onset diabetes. By analyzing patient data across multiple healthcare domains, including hospital diagnoses, lab measurements, medications, and primary care visits, this project seeks to identify subtle symptoms and risk factors of the cancer early on and to pinpoint high-risk individuals. The project aims to make significant strides in personalized cancer detection through timely diagnosis and individual risk profiles.

Ovarian cancer and early symptoms

Ovarian cancer is the 8th most common cancer among women worldwide, often going undiagnosed until its later stages due to vague symptoms that mimic milder conditions or menopause. As a result, survival rates remain low, with only a 50% chance of survival over five years. However, when detected at an early stage (stage 1), this rate can increase to 90%.
In this project we seek to improve understanding and detection of early symptoms of ovarian cancer using multimodal health data and explainable AI methods.