Computer Assisted Pit-Pattern Classification Using Wavelet techniques

According to the American Cancer Society colon cancer ist the third most common type of cancer in males and fourth in females in western countries. Therefore screening programs have been established worldwide in order to detect lesions with a malignant potential or early cancer. The current goldstandard for the detection of lesions in the colon is flexible videoendoscopy. The colonoscope allows a physician or surgeon to examine the colon and the distal part of the ileum and to take tissue samples. Furthermore, polypoid and flat lesions can be resected endoscopically, therefore leading to reduced incidence of colonic carcinoma in populations screened using colonoscopy. Modern colonoscopes are using high resolution CCD and storage of pictures and video sequences can be done using additional hardware. Pictures can be used for quality assurance or for later review. Another possibility arising from the ability of taking pictures from the colon is analysis of images or video sequences with the assistance of computers. This allows computer assisted detection and classification of lesions. Recently, magnification endoscopy has been introduced into the market. These special "zoom-endoscopes" provide images with magnification factors up to 150x and higher. Current methods include digital and optical zooms, using adjustable lenses. Additionally, dye spraying is used in order to enhance the contrast of the picture (chromoendoscopy). In the colon the most common dye used is indigo carmine, alternatively, methylene blue is used by some centers. The combination of chromo- and zoom-endoscopy leads to characteristic mucosal surface patterns which can be interpreted by an experienced examiner. However, difficulties with respect to result accuracy and also inter-examiner variability have been reporteted recently. Consequently, this project proposal aims at developing techniques for a computer assisted clinical assessment employing and comparing different types of wavelet-based technique for an automated classification of visual data aquired by a magnifying colonoscope corresponding to different types of lesions. In preparatory work we have found that techniques developed for wavelet-based texture classification lead to reasonable classification results, however, the accurracy is too low to allow for clinical use. We will develop and use wavelet-based techniques specifically tailored to the characteristc properties of pit patterns in this project.

According to the American Cancer Society colon cancer ist the third most common type of cancer in males and fourth in females in western countries. Therefore screening programs have been established worldwide in order to detect lesions with a malignant potential or early cancer. The current goldstandard for the detection of lesions in the colon is flexible videoendoscopy. The colonoscope allows a physician or surgeon to examine the colon and the distal part of the ileum and to take tissue samples. Furthermore, polypoid and flat lesions can be resected endoscopically, therefore leading to reduced incidence of colonic carcinoma in populations screened using colonoscopy. Modern colonoscopes are using high resolution CCD and storage of pictures and video sequences can be done using additional hardware. Pictures can be used for quality assurance or for later review. Another possibility arising from the ability of taking pictures from the colon is analysis of images or video sequences with the assistance of computers. This allows computer assisted detection and classification of lesions. Recently, magnification endoscopy has been introduced into the market. These special `zoom-endoscopes` provide images with magnification factors up to 150x and higher. Current methods include digital and optical zooms, using adjustable lenses. Additionally, dye spraying is used in order to enhance the contrast of the picture (chromoendoscopy). In the colon the most common dye used is indigo carmine, alternatively, methylene blue is used by some centers. The combination of chromo- and zoom-endoscopy leads to characteristic mucosal surface patterns which can be interpreted by an experienced examiner. However, difficulties with respect to result accuracy and also inter-examiner variability have been reporteted recently. Consequently, this project aimed at developing techniques for a computer assisted clinical assessment employing and comparing different types of wavelet-based technique for an automated classification of visual data aquired by a magnifying colonoscope corresponding to different types of lesions. The developed feature extraction and classification technique achieved very high accuracy in distinguishing between neoplastic and non-neoplastic colonic lesions. In particular, results from other endoscopic imaging modalities have been inferior to the results found in this project: narrow-band imaging (NBI), confocal endomicroscopic imaging and white-light colonoscopic imaging were reported to deliver lower classification accuracy as compared to the approach used in this project.