Biomedical Signal Processing
 
Biomedical Signal Processing
Statistical and Bio-Inspired Machine Learning
Data Mining
Audio, Speech and Language Processing
 
 
Items
Unsupervised classification and vector quantization methods applied to renal functional segmentation
 
  
> MALIS Home > Research Topics > Machine Learning > Biomedical Signal Processing > Unsupervised classification and vector quantization methods applied to renal functional segmentation

MAchine Learning and Interactive Systems

Unsupervised classification and vector quantization methods applied to renal functional segmentation
 
  by Chevaillier Beatrice
 
 

Among the magnetic resonance imaging (MRI) techniques aiming at renal functional study, dynamic contrast-enhanced MRI (DCE-MRI) with gadolinium chelates injection is the most widely used. Several parameters can be non-invasevely extracted from time-intensity curves of different anatomical compartments. Segmentation of internal kidney structures is thus crucial for functional assessment. Manual segmentation by a radiologist is fairly delicate because images are blurred and highly noisy. Moreover the different compartments are not visible during the same perfusion phase because of contrast changes ; consequently they cannot be delineated on a single image. Radiologists have to select the few most suitable frames : the operation is time-consuming and misregistration or through-plane motion can lead to great variations in functional analysis. Nevertheless the differences between temporal evolution of contrast in each anatomical region can be used to perform functional segmentation. Pixels can thus be classified according to their time-intensity curves using some unsupervised classification method or vector quantization algorithm followed by a semi-automated merging step. These methods may offer more robustness and reproducibility because the whole sequence (about 200 to 300 frames) is used instead of only a few frames for manual segmentation. Generated time saving is also considerable : manual segmentation requires 12 to 15 minutes, versus about 20 seconds for the semi-automated methods.