Improving hypoxia map estimation by using model-free classification techniques in DCE-MRI images

The vascular microenvironment of tumors is a key determinant of the tumor pathophysiology. Hypoxia, i.e. lack of sufficient oxygen supply, might affect significantly the treatment efficacy of solid tumors making it an important imaging biomarker. The ability to characterize oxygen perfusion of the tumor can provide prognostic information about the tumor progression and risk of metastases. In this work, Dynamic Contrast Enhanced Magnetic Resonance Imaging (DCE-MRI), a non-invasive method, has been used for the detection of tumor hypoxic areas on neck sarcoma data. Data analysis was performed using a pattern recognition (PR) technique able to automatically identify potential tumor hypoxic regions along with a well-established pharmacokinetic (PK) model for computing perfusion parameters. The paper presents a novel method for the initialization of the PR technique through realistic assumptions in order to overcome instability issues found in random initialization. To this end, the PR technique was initialized using two novel approaches based on the wash-in part of the dynamic acquisition and the k
^trans
map derived from the PK analysis. The results, from these different implementations show high correlation between them and consistently lead to the separation of the tumor area into well-perfused, hypoxic and necrotic regions.