Quantitative imaging for automated brain tumour diagnosis

In Belgium, 800 people are diagnosed with a primary brain tumour each year. Their life expectancy depends strongly on the tumour type, and ranges from several decades for low-grade tumours to only about one year for the most malignant types. An accurate pathological diagnosis is therefore of primary importance. In clinical practice the workflow from initial consultation to therapy involves many steps within a multidisciplinary team of radiologists, neurosurgeons, neuropathologists, nuclear medicine professionals, radiotherapists, etc. In each step of this pipeline, there is a degree of subjectivity involved in the decision process. However, in literature there is growing evidence that a quantitative, and thus objective, approach can yield a correct diagnosis and therapy prediction. The goal of this project is therefore to investigate how a quantitative analysis of medical images (both MRI and PET) can contribute to a personalised medicine. In this project, we will implement a method for automated brain tumour diagnosis, using textural analysis in combination with other quantitative tumour parameters, such as volume, shape and histogram features. Using advanced classification algorithms, such as Support Vector Machines or Convolutional Neural Networks, an algorithm for automated diagnosis will be implemented.

Researcher: Stijn Bonte