Room: AAPM ePoster Library
Purpose: To perform pattern recognition using non-negative matrix factorization (NMF) for spatiotemporal analysis in dynamic contrast-enhanced (DCE) MRI. This approach shows the heterogenous distribution of perfusion in cancer tumours, and enables longitudinal assessment of response to radiation therapy in high grade soft-tissue sarcoma.
Methods: Data in this study was collected from patients (n=18) with histologically-proven high-grade soft tissue sarcoma. Three MRI exams were performed over the course of neoadjuvant radiotherapy (before, during, and after 50 Gy/25 fractions), under institutional approval and informed consent. Imaging was performed on a 1.5 T MRI scanner (Signa, GE Healthcare). Dynamic T1-weighted time-series images were acquired during intravenous injection of gadobutrol. Manual contours of the sarcoma, produced by a radiation oncologist, were used to exclude healthy tissue from the analysis. NMF was used to identify two (k=2) time-course patterns in the image data (the sources) and to produce the corresponding spatial distributions (the weight maps). A multi-run-averaged NMF approach was used to produce the most reliable sources and weight maps for each patient.
Results: NMF identified source curves that resemble signal enhancement curves, described here as high perfusion and low perfusion. These sources showed very strong similarities across patients and timepoints. The weight maps corresponding to high and low perfusion curves reveal heterogeneity in tumour perfusion, which changes across timepoints.
Conclusion: Pattern recognition algorithms can identify perfusion curves in DCE-MRI data and produce maps that reflect the heterogenous distribution of these perfusion patterns in high-grade soft-tissue sarcoma. NMF approaches with run averaging could be a practical and robust data-driven approach to characterizing blood supply in tumours. This approach could be used to study tumour progression throughout the course of radiotherapy treatments. Correlation of NMF tumour perfusion metrics to clinical outcomes is currently underway.
Pattern Recognition, Perfusion Imaging