Comparison of two algorithms for the application of real-time image mosaicking in neuroendoscopy.

OBJECT: Neuroendoscopy is used more and more frequently in neurosurgical procedures and has become an important tool in the neurosurgical armamentarium. However, the main restriction of neuroendoscopy is the limited field of view. A better overview of the area of interest would increase surgical safety and decrease procedure-related morbidity rates. In the present study, the authors aimed to improve this restriction by using and comparing two algorithms to create endoscopic panoramic images, which increase the field of view during neuroendoscopic procedures. METHODS: Different endoscopic methods with or without a stand and with linear or circular endoscope movements were performed in cadaveric ventricles. Video of the endoscopy was used to create image mosaics of the lateral ventricle with the help of the Kourogi or LogSearch (LS) algorithm. In the LS algorithm, different template sizes were used. Three observers graded the quality of the image mosaic in terms of usefulness in surgery. The fastest frame rate was 3-4 frames/second. RESULTS: The LS algorithm with a larger template size showed significantly better results for the creation of image mosaics than the Kourogi algorithm in linear endoscopic movement with or without a stand. In circular endoscopic movements, the results seemed to be better with the LS algorithm but were not significantly different from those obtained with the Kourogi algorithm. In summary, image quality in the experimental paradigms was satisfying. CONCLUSIONS: Results in the study showed that the creation of image mosaics is possible and reliable with the featured algorithms. Image mosaicking is an applicable device for neuroendoscopy and can increase the field of view during endoscopic procedures. Its use can increase the safety and the field of application of neuroendoscopy. However, faster frame rates will be required to create a smooth image for practical use during surgery.

On the performance of neuronal matching algorithms.

For a solution of the visual correspondence problem we have modified the Self Organizing Map (SOM) to map image planes onto another in a neighborhood- and feature-preserving way. We have investigated the convergence speed of this SOM and Dynamic Link Matching (DLM) on a benchmark problem for the solution of which both algorithms are good candidates. We show that even after careful parameter adjustment the SOM needs a large number of simple update steps and DLM a small number of complicated ones. The results are consistent with an exponential vs. polynomial scaling behavior with increased pattern size. Finally, we present and motivate a rule for adjusting the parameters of DLM for all problem sizes, which we could not find for SOM.