Abstract
We present a parallel matching component of an integrated system for the automatic analysis of MRI-breast images towards the early detection of breast cancer. The system operates on taken images using the method of dynamic contrast-enhanced MRI. Suspicious breast lesions are automatically marked with colors, thus directing the physician's attention towards the critical regions. A proper and careful decision procedure is needed to differentiate between increases of signal intensity triggered by noise and tissue dislocations (motion artifacts) and increases that are triggered by an accumulation of contrast agent in the related breast region. We present our component for image matching using self-organizing maps (SOM), which enables the system to work properly even with image sequences that are strongly deformed by the patients breathing movements. To reach the time constraint of 15 minutes in medical practice we decide to implement a parallel architecture for the neural network matcher, which works on all computers in the heterogeneous network of our medical partners. The system is tested on real patient data and is now being refined in cooperation with our partner hospital for Radiology and Nuclear Medicine in Mainz.