ANNA – Artificial Neural Network Analyzer
A Framework for the Automatic Inspection of TRUS Images
The project ANNA (Artificial Neural Network Analyzer) aims at the design of a framework to analyze ultrasound images by means of signal processing combined with methods from the field of artificial intelligence (neural networks, self-organizing maps, etc.). Although not obvious ultrasound images do contain information that cannot be recognized by the human visual system and that do provide information about the underlying tissue. On the other hand the human visual system recognizes virtual structures in ultrasound images that are not related at all to the underlying tissue. Especially interesting in this regard is the fact that the careful combination several texture descriptor based filters is suited for an analysis by artificial neural networks and that suspicious regions can be marked reliably. The specific aim of the framework is to automatically analyze conventional rectal ultrasound (TRUS) images of the prostate in order to detect suspicious regions that are likely to relate to a primary cancer focus. These regions are marked for a subsequent biopsy procedure. The advantages of such an analysis are the significantly reduced number of biopsies compared to a random or systematic biopsy procedure to detect a primary cancer and the significantly enhanced success rate to extract primary cancer tissue with a single biopsy procedure. On one hand this results in a faster and more reliable diagnosis with significantly decreased intra-examiner variability, on the other hand the discomfort of the patient due to multiple biopsy sessions is dramatically reduced.