Distributed Learning

 Advanced medical prognostic and predictive techniques require vast amounts of data to achieve acceptable accuracy. At the same time health care centres are bound by legal and ethical reason not to share patient data with large databases. Distributed Learning solves the big data vs data privacy dilemma by taking the prognostic and predictive models to each centre and training (“learning”) them on-site. 


Clinical trials encompass only a small percentage of the patient population. In order to improve healthcare using advanced and personalised disease models, larger patient databases are necessary, but patient anonymity needs to be preserved. The euroCAT project aims to advance personalised and participatory medicine by connecting various medical institutions through distributed learning.


Every year millions of medical images are recorded and stored in medical databases. Most of them are evaluated by an expert in the process of disease classification and ensuing therapy. Radiomics turns medical images into mineable data by extracting hundreds or even thousands of features that are unique to to each image, allowing those to be correlated with clinically relevant data such as tumour type or prognosis. Using these correlations medical experts have another tool in their belt to help fight disease.

Virtual Spacer

Because the rectum is so close to the prostate, when prostate cancer is treated with radiotherapy the rectum may be damaged by the otherwise healing radiation. One way to minimise this effect is to increase the distance between the two organs by placing a “spacer” between them. A computer modelled “virtue spacer”, based on a patient CT scan, is able to asses the how well this very invasive technique will work on an individual level, saving many patients from this technique.

Virtual Patient Avatar

When going from one medical centre to another, or moving towns or even countries, it is often very hard to keep track of one’s own medical records, or have them transferred. And even when this happens, the information is only processed by the experts in small chunks. With the Virtual Patient Avatar, you can have all this information on your phone, and also recording habits such as exercise and diet, which together with sophisticated models can calculate the your personal probabilities for multiple diseases.

Decision Support

A significant minority of patients undergoing radiotherapy is more prone to suffering severe side-effects due to radiation toxicity. Through a simple test of mitochondrial DNA from a saliva sample we can advise these patients to undergo the very costly but less severe proton therapy, while safely increasing the cancer killing radiation for those patients with higher radiation resistance.

Hypoxia Activated Pro-drugs

Hypoxic cells are cells which receive too little oxygen. Research has shown that hypoxic tumour cells are more malignant and more resistant to conventional therapies. Hypoxia activated pro-drugs solve this problem elegantly by only activating in such environments, yet clinical trials have shown these not to be efficient. This is mainly due to patient selection, as it turns out that hypoxia alone is not a sufficient indicator for these drugs to work. We have developed a model to pre-select patients for this treatment and increase the efficacy for this sub-group.

Individualized Patient Decision Aids

©2017 by D-Lab