Segmentation-free scheme for computer-assisted image interpretation: application to CT Colonography

F. Chandelier, T. Cabrera, P. Kocsis, L. Stein, V. Demers; ECR 2013 Book of Abstracts (B-0619)
Purpose: To present a paradigm shift in computer-assisted image interpretation and to assess its application in CT colonography. A novel segmentation-free scheme has been developed with the hypothesis that accurate colon segmentation is neither necessary nor an end clinical objective for delivering efficient support such as colon centreline, electronic colon cleansing (ECC) and computer-aided detection of lesions (CAD).

Methods and Materials: Thick regions are expanded until they encompass statistically significant representations of the colon wall and its surroundings (Air, Tag, Soft/Hard tissue). From these regions, a coarse colon centreline is inferred through topological analysis guided by anatomic constraints; ECC is obtained by determining a transposing function depicting the distribution of voxels in a given Air-region, and applying said function to Tag-regions depicting statistically similar colon wall surroundings; unlike surface shape analysis, a CAD identifies local hyper-density concentrations from implicit surfaces flux, at a given depth related to the targeted lesion size. The complete schemes were retrospectively tested on 111 and 135 patients datasets for {centreline; electronic cleansing} and CAD, respectively.

Results: Colon centreline extraction was accurate in 85% of combined cathartic and tagged datasets. CAD performance was similar to that in the literature, specifically a sensitivity of 80% [70-85] for lesions at least 7 mm featuring a false-positive rate less than 5 per-patient.

Conclusion: The presented novel paradigm in computer-assisted image interpretation prevents traditional limitations arising from "accurately segmenting" a specific organ for further processing. To our best knowledge, this is the first attempt to design and validate such a paradigm shift in CT colonography.