BACKGROUND AND AIMS
Probe-based confocal laser endomicroscopy (pCLE; Cellvizio, Mauna Kea Technologies, Paris, France) enables in vivo histology during colonoscopy and may allow endoscopists to make real-time diagnoses. A collaboration of five experts proposed a new pCLE classification for colonic use. The aim of this study was to assess interobserver agreement and accuracy of the new pCLE classification in the colon.
PATIENTS AND METHODS
Eligible patients were prospectively investigated by pCLE. A subset of 13 pCLE video sequences was reviewed post hoc for the establishment of a new classification, which comprised three vessel categories and seven crypt categories. All five blinded observers then scored another set of 102 video sequences, using the new classification. Histopathology was used as a reference standard.
RESULTS
The interobserver agreements on vessel and crypt architecture were 'fair' with kappa values of 0.29 and 0.27, respectively. When the classification was reduced to neoplasia vs. non-neoplasia (i.e. vessel or crypt type 3), overall agreement became 'moderate' (κ = 0.56). Overall sensitivity and specificity for predicting neoplasia was 66 % and 83 %, respectively. When all observers agreed (69 % of videos), the corresponding figures became 80 % and 95 %.
CONCLUSION
A new classification for pCLE in the colon had a 'moderate' interobserver agreement for differentiating neoplasia from non-neoplastic tissue in the colon. The overall accuracy (81 %) for predicting neoplasia was acceptable and became excellent (94 %) when all five observers agreed. Future research should focus on refinement and validation of the classification.