Fusing in vivo and ex vivo NMR sources of information for brain tumor classification
Croitor-Sava, A R; Martinez-Bisbal, M C; Laudadio, T; Piquer, J; Celda, B; Heerschap, A; Sima, D M; Van Huffel, S; Croitor-Sava, A R; Department of Electrical Engineering, ESAT-SCD, Katholieke Universiteit Leuven, Leuven-Heverlee, Belgium; IBBT-KU Leuven Future Health Department, Leuven, Belgium; Martinez-Bisbal, M C; CIBER of Bioengineering, Biomaterials and Nanomedicine, ISC-III, Spain; Departamento de Química-Física, Facultad de Química, Universidad de València, València, Spain; Laudadio, T; Department of Electrical Engineering, ESAT-SCD, Katholieke Universiteit Leuven, Leuven-Heverlee, Belgium; Istituto per le Applicazioni del Calcolo ‘M Picone’, National Research Council, (IAC-CNR), Bari, Italy; Piquer, J; Neurosurgery Service, Hospital de La Ribera, Carretero Alzira-Corbera, Valencia, Spain; Celda, B; CIBER of Bioengineering, Biomaterials and Nanomedicine, ISC-III, Spain; Departamento de Química-Física, Facultad de Química, Universidad de València, València, Spain; Heerschap, A; Department of Neurosurgery, University of Nijmegen, University Medical Center, Nijmegen, The Netherlands; Sima, D M; Department of Electrical Engineering, ESAT-SCD, Katholieke Universiteit Leuven, Leuven-Heverlee, Belgium; IBBT-KU Leuven Future Health Department, Leuven, Belgium; Van Huffel, S; Department of Electrical Engineering, ESAT-SCD, Katholieke Universiteit Leuven, Leuven-Heverlee, Belgium; IBBT-KU Leuven Future Health Department, Leuven, Belgium
Журнал:
Measurement Science and Technology
Дата:
2011-11-01
Аннотация:
In this study we classify short echo-time brain magnetic resonance spectroscopic imaging (MRSI) data by applying a model-based canonical correlation analyses algorithm and by using, as prior knowledge, multimodal sources of information coming from high-resolution magic angle spinning (HR-MAS), MRSI and magnetic resonance imaging. The potential and limitations of fusing in vivo and ex vivo nuclear magnetic resonance sources to detect brain tumors is investigated. We present various modalities for multimodal data fusion, study the effect and the impact of using multimodal information for classifying MRSI brain glial tumors data and analyze which parameters influence the classification results by means of extensive simulation and in vivo studies. Special attention is drawn to the possibility of considering HR-MAS data as a complementary dataset when dealing with a lack of MRSI data needed to build a classifier. Results show that HR-MAS information can have added value in the process of classifying MRSI data.
924.7Кб