Computer-aided classification of colorectal polyps based on vascular patterns: a pilot study

Abstract

Automated classification of colonic polyps on the basis of NBI vascularization features is feasible, but classification by observers is still superior. Further research is needed to clarify whether the performance of the automated classification system can be improved.

Predicting Pathology in Medical Decision Support Systems in Endoscopy of the Gastrointestinal Tract

“…Although, at a first glance, the approach presented in (Häfner et al, 2010b) seems to perform better in terms of the overall prediction accuracy compared to (Tischendorf et al, 2010), a comparison of the overall system accuracies would be meaningless. This is mainly due to the fact that both approaches are based on quite different image databases.…”

“…• Number of images used Comparing the total number of images between the two approaches clearly shows that the number of images used in (Häfner et al, 2010b) is roughly three times higher compared to (Tischendorf et al, 2010). Since, as pointed out above, we have no figures about the number of patients in case of (Häfner et al, 2010b), it is not possible to assess whether this is beneficial (since, in case of only a few patients, this would lead to overfitting).…”

“…In order to compare two classifier-based approaches found in literature we picked two recently published approaches (found in Tischendorf et al (2010) and Häfner et al (2010b)). Some facts about these approaches are listed in Table 1 While both methods are similar in terms of the GI tract part under investigation, the pathology of interest, the prediction goal (hence, also the same number of classes), the cross-validation protocol used, and the way the ground truth has been obtained, we also notice some differences between these works.…”

“…While in (Tischendorf et al, 2010) NBI has been used, the approach presented in (Häfner et al, 2010b) is based on the application of topical staining (using color dyes), which still is common practice in clinical routine.…”

“…• Number of patients While in (Tischendorf et al, 2010) the experiments are based on images originating from 223 patients, the work presented in (Häfner et al, 2010b) makes no statements about this detail. Nevertheless, as already pointed out earlier, using too many images from one patient may lead to an overfitting.…”